Method for determining automatic driving control data and method for training machine learning model
By employing a multi-branch machine learning model training method, the problem of inaccurate trajectory error prediction for autonomous vehicles in different scenarios was solved, thereby improving safety and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING BAIDU NETCOM SCI & TECH CO LTD
- Filing Date
- 2023-01-18
- Publication Date
- 2026-07-10
AI Technical Summary
Existing autonomous driving technologies struggle to accurately predict trajectory errors in various scenarios, leading to safety and efficiency issues, especially in high-curvature turning scenarios where the risk of collisions with obstacles is high.
A multi-branch machine learning model, consisting of N first learners and M second learners, is used. The model parameters are adjusted through a target loss function to train a target machine learning model for accurately predicting trajectory errors.
It improves the accuracy of trajectory error prediction and enhances the driving safety and traffic efficiency of autonomous vehicles in different scenarios.
Smart Images
Figure CN116142215B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of computer technology, and more particularly to the fields of artificial intelligence, autonomous driving, machine learning and intelligent transportation technology. Specifically, it relates to a method for determining autonomous driving control data, a method for training machine learning models, an apparatus, a device, a storage medium, a program product and an autonomous vehicle. Background Technology
[0002] Autonomous driving is an application branch of artificial intelligence technology, and how to control vehicle driving safely and efficiently is a major challenge we face today.
[0003] To achieve more intelligent autonomous driving, targeted optimizations are needed based on the characteristics of different scenarios to ensure the safety of autonomous vehicles. For example, in scenarios where autonomous vehicles are traveling straight, the error between their actual and planned trajectories is relatively small; only obstacles that overlap with the planned trajectory pose a potential collision risk. However, in scenarios with sharp turns, the actual and planned trajectories deviate significantly. This error introduces uncertainty into collision detection based on the planned trajectory, making the risk of collisions with obstacles higher and the safety lower in such scenarios. Summary of the Invention
[0004] This disclosure provides a method for determining autonomous driving control data, a method for training machine learning models, an apparatus, a device, a storage medium, a program product, and an autonomous vehicle.
[0005] According to one aspect of this disclosure, an autonomous driving control data determination method is provided, comprising: processing target steering control parameter data for a target vehicle using a target machine learning model to obtain trajectory error data for the target vehicle; determining target control data for controlling the driving of the target vehicle based on the trajectory error data, wherein the target machine learning model is obtained by: acquiring target steering control parameter data samples for the target vehicle and an initial machine learning model, wherein the initial machine learning model includes at least one branch, each branch including N first learners and M second learners, the N first learners being distinct, N being an integer greater than or equal to 2, and M being a positive integer less than N; inputting the target steering control parameter data samples into each first learner of the initial machine learning model to obtain N first output results; determining M second output results based on the first output results and the second learners; determining feedback values based on the M second output results and a target loss function; and adjusting the model parameters of the initial machine learning model based on the feedback values to obtain the target machine learning model.
[0006] According to another aspect of this disclosure, a machine learning model training method is provided, comprising: acquiring target steering control parameter data samples for a target vehicle and an initial machine learning model, wherein the initial machine learning model includes at least one branch, each branch including N first learners and M second learners, the N first learners being distinct from each other, N being an integer greater than or equal to 2, and M being a positive integer less than N; inputting the target steering control parameter data samples into each first learner of the initial machine learning model to obtain N first output results; determining M second output results based on the first output results and the second learners; determining feedback values based on the M second output results and a target loss function; and adjusting the model parameters of the initial machine learning model based on the feedback values to obtain a target machine learning model.
[0007] According to another aspect of this disclosure, an autonomous driving control data determination device is provided, comprising: a trajectory error data determination module, configured to process target steering control parameter data for a target vehicle using a target machine learning model to obtain trajectory error data for the target vehicle; and a target control data determination module, configured to determine target control data for controlling the driving of the target vehicle based on the trajectory error data, wherein the target machine learning model is obtained using the following module: a first acquisition module, configured to acquire target steering control parameter data samples for the target vehicle and an initial machine learning model, wherein the initial machine learning model includes at least one branch, and each branch includes N... There are N first learners and M second learners, where the N first learners are distinct, N is an integer greater than or equal to 2, and M is a positive integer less than N; a first output result determination module is used to input the target steering control parameter data sample into each of the first learners of the initial machine learning model to obtain N first output results; a second output result determination module is used to determine M second output results based on the first output results and the second learners; a feedback value determination module is used to determine the feedback value based on the M second output results and the target loss function; and a target machine learning model determination module is used to adjust the model parameters of the initial machine learning model based on the feedback value to obtain the target machine learning model.
[0008] According to another aspect of this disclosure, a machine learning model training apparatus is provided, comprising: a first acquisition module, configured to acquire target steering control parameter data samples for a target vehicle and an initial machine learning model, wherein the initial machine learning model includes at least one branch, each branch including N first learners and M second learners, the N first learners being distinct from each other, N being an integer greater than or equal to 2, and M being a positive integer less than N; a first output result determination module, configured to input the target steering control parameter data samples into each first learner of the initial machine learning model to obtain N first output results; a second output result determination module, configured to determine M second output results based on the first output results and the second learners; a feedback value determination module, configured to determine feedback values based on the M second output results and a target loss function; and a target machine learning model determination module, configured to adjust the model parameters of the initial machine learning model based on the feedback values to obtain a target machine learning model.
[0009] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the methods of embodiments of this disclosure.
[0010] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods of embodiments of this disclosure.
[0011] According to another aspect of this disclosure, a computer program product is provided, including a computer program stored on at least one of a readable storage medium and an electronic device, wherein the computer program, when executed by a processor, implements the methods of embodiments of this disclosure.
[0012] According to another aspect of this disclosure, an autonomous vehicle is provided, including electronic devices, wherein the target vehicle drives according to target control data.
[0013] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description
[0014] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:
[0015] Figure 1This schematically illustrates a system architecture diagram of a method for determining autonomous driving control data, a machine learning model training method, and an apparatus according to embodiments of the present disclosure.
[0016] Figure 2 A flowchart illustrating a machine learning model training method according to an embodiment of the present disclosure is shown schematically.
[0017] Figure 3 A schematic diagram illustrating a machine learning model training method according to another embodiment of the present disclosure is shown.
[0018] Figure 4 A schematic diagram illustrating a method for determining autonomous driving control data according to an embodiment of the present disclosure is shown.
[0019] Figure 5 A block diagram of an apparatus for determining autonomous driving control data according to an embodiment of the present disclosure is shown schematically.
[0020] Figure 6 A block diagram of a machine learning model training apparatus according to an embodiment of the present disclosure is schematically shown; and
[0021] Figure 7 A block diagram of an electronic device that can implement the method for determining autonomous driving control data and the method for training machine learning models according to embodiments of the present disclosure is illustrated schematically. Detailed Implementation
[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.
[0024] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.
[0025] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).
[0026] Autonomous driving is an application branch of artificial intelligence technology, and how to control vehicle driving safely and efficiently is a major challenge we face today.
[0027] To achieve more intelligent autonomous driving, ensuring the safety of autonomous vehicles is crucial. Autonomous driving needs to cover all scenarios, requiring targeted optimization based on the characteristics of different scenarios. For example, collision detection differs between straight-ahead and turning scenarios for autonomous vehicles. In straight-ahead scenarios, autonomous vehicles can closely follow the planned trajectory, meaning the actual trajectory deviates little from the planned trajectory. In this case, only obstacles overlapping with the planned trajectory pose a potential collision risk. However, in scenarios with sharp turns, the actual trajectory deviates significantly from the planned trajectory. This error introduces uncertainty into collision detection, meaning obstacles not overlapping with the planned trajectory may also collide with the autonomous vehicle. This makes collisions more risky and less safe for autonomous vehicles in sharp turns.
[0028] Therefore, it is necessary to accurately predict the trajectory error of autonomous vehicles. The trajectory error can be used to accurately predict whether autonomous vehicles are at risk of colliding with obstacles in turning scenarios, thereby ensuring the traffic safety and efficiency of autonomous vehicles.
[0029] Some implementations use fixed or variable buffer distances to increase safety margins and compensate for errors between the autonomous vehicle and the planned trajectory to avoid collisions. While this approach can improve the driving safety of autonomous vehicles, it also reduces the diversity of planned trajectories. In extreme cases, it may lead to the inability to generate a feasible planned trajectory, or the autonomous vehicle's trajectory planning may fail or stall, affecting traffic efficiency. For example, since there is a certain deviation between the actual driving trajectory and the planned trajectory, and the magnitude of the trajectory error cannot be determined in advance, a safety buffer is pre-set, such as maintaining a distance of 1 meter between the planned trajectory and obstacles. This safety buffer is set based on human experience and cannot predict the specific value of the actual error. If the manually set safety buffer is too small, a collision may occur. If the manually set safety buffer is too large, a path (planned trajectory) may not be successfully planned. For example, when turning, there are obstacles on both sides, and the gap between the two obstacles is just wide enough for the autonomous vehicle to pass under its current width. However, because the safety buffer is set too large, the planning module (e.g., using the planning module to determine the planned trajectory) may conclude that the autonomous vehicle will collide with the obstacles and cannot pass.
[0030] Figure 1 The illustration schematically depicts the system architecture of a method for determining autonomous driving control data, a machine learning model training method, and an apparatus according to an embodiment of the present disclosure. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.
[0031] like Figure 1 As shown, the system architecture 100 in this embodiment may include: a terminal 101 for sample generation, a terminal 102 for machine learning model training, a terminal 103 for determining target control data, a network 104, and vehicles 105, 106, and 107.
[0032] In this embodiment, terminal 101 can be used to generate target steering control parameter samples or initial steering control parameter samples. Terminal 102 can execute a corresponding machine learning model training method based on the target steering control parameter samples obtained by terminal 101 to train a target machine learning model. Terminal 103 can process the target steering control parameter data based on the target machine learning model obtained by terminal 102 to obtain target control data for controlling the driving of the target vehicle. The target control data for controlling the driving of the target vehicle obtained by terminal 103 can be sent to vehicles 105, 106, and 107, which then control the vehicle's driving according to the target control data.
[0033] It should be noted that determining autonomous driving control data, training machine learning models, and generating samples can be done on the same terminal or on different terminals.
[0034] Terminal 101, Terminal 102 and Terminal 103 can be servers or server clusters.
[0035] Network 104 is a medium used to provide communication links between terminals 101, 102, 103 and vehicles 105, 106, 107. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.
[0036] In one embodiment, vehicles 105, 106, and 107 can interact with terminal 103. For example, vehicles 105, 106, and 107 can transmit target steering control parameter data to server 104. Terminal 103 then determines target control data for controlling the movement of the target vehicles based on the target steering control parameter data. This target control data can, for example, be used to control the vehicle's chassis drive-by-wire system.
[0037] In another example, vehicles 105, 106, and 107 can perform data processing. The vehicle's infotainment system can have data processing capabilities, allowing it to determine target control data for controlling the target vehicle's movement based on the target steering control parameters.
[0038] For example, the vehicle includes electronic devices, which include, but are not limited to, vehicle infotainment systems. These electronic devices can execute the autonomous driving control data determination method according to embodiments of this disclosure. The vehicle infotainment system may include a chassis drive-by-wire system.
[0039] Vehicles 105, 106, and 107 can be autonomous vehicles (target vehicles).
[0040] It should be understood that Figure 1The number of vehicles, networks, and terminals shown is merely illustrative. Any number of vehicles, networks, and terminals can be included depending on implementation needs.
[0041] It should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0042] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.
[0043] This disclosure provides a method for training a machine learning model, which will be described below in conjunction with... Figure 1 The system architecture, referencing Figures 2-3 This describes a machine learning model training method according to exemplary embodiments of the present disclosure. The machine learning model training method of the embodiments of the present disclosure may, for example, be derived from... Figure 1 The terminal 102 shown is used to execute this.
[0044] Figure 2 A flowchart illustrating a machine learning model training method according to an embodiment of the present disclosure is shown schematically.
[0045] like Figure 2 As shown, the machine learning model training method 200 of this embodiment may include, for example, operations S210 to S250.
[0046] In operation S210, target steering control parameter samples and an initial machine learning model for the target vehicle are obtained.
[0047] The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners.
[0048] The N first learners are all distinct, where N is an integer greater than or equal to 2 and M is a positive integer less than N.
[0049] The initial machine learning model has initial model parameters, which may include, for example, the model parameters of the first learner and the model parameters of the second learner.
[0050] A learner can be understood as a single machine learning model.
[0051] The target steering control parameter samples can be understood as training samples of the parameter data used to control the steering of the target vehicle. These samples are labeled, indicating the actual trajectory error data. The trajectory error data can be understood as the error between the actual trajectory and the planned trajectory of the target vehicle. It should be noted that the target steering control parameter samples are training samples corresponding to the parameter data used to control the steering of the target vehicle, and the corresponding trajectory error data is the error between the actual trajectory and the planned trajectory of the target vehicle in steering, lateral control, and cornering scenarios.
[0052] In operation S220, the target is shifted to each first learner of the initial machine learning model by inputting the control parameter samples, resulting in N first output results.
[0053] In operation S230, based on the first output result and the second learner, M second output results are determined.
[0054] The second output may include trajectory error data predicted by the current initial machine learning model.
[0055] In operation S240, the feedback value is determined based on the M second output results and the target loss function.
[0056] The objective loss function can be used to evaluate the performance of an initial machine learning model in predicting trajectory errors. Feedback values, for example, can evaluate the accuracy of the current initial machine learning model in predicting trajectory errors.
[0057] In operation S250, the model parameters of the initial machine learning model are adjusted based on the feedback values to obtain the target machine learning model.
[0058] The target machine learning model is used to determine trajectory error data for the target vehicle.
[0059] The target machine learning model has the same model structure as the initial machine learning model but different model parameters. The target machine learning model still includes at least one branch, and each branch includes N first learners and M second learners.
[0060] According to the machine learning model training method of this disclosure, by inputting target steering control parameter samples into each first learner of an initial machine learning model, N first output results are obtained, and M second output results determined by the first output results and the second learner can be used to characterize the trajectory error prediction results of the initial machine learning model for the target steering control parameter samples. By adjusting the model parameters of the initial machine learning model according to the feedback values, the performance of the obtained target machine learning model is better, and the accuracy of trajectory error prediction using the target machine learning model is higher.
[0061] According to the machine learning model training method of this disclosure, the target machine learning model obtained through the above operations still includes multiple branches, each of which can predict trajectory error data. The multiple branches can be combined to obtain more accurate trajectory error data. Furthermore, each branch includes N first learners and M second learners, allowing the N first learners and M second learners to learn a wider range of features related to the predicted trajectory error data, achieving ensemble learning. This results in a more accurate prediction of trajectory error data based on target steering parameter data, adapting to scenarios such as high-curvature steering, lateral control, and curves, thereby improving the driving safety and traffic efficiency of the target vehicle.
[0062] For example, the machine learning model training method according to another embodiment of this disclosure may further include: acquiring initial steering control parameter samples for a target vehicle; and determining target steering control parameter data based on correlation evaluation information among multiple candidate steering control parameter data.
[0063] The initial steering control parameter sample includes multiple candidate steering control parameter data related to vehicle steering. The target steering control parameter sample includes target steering control parameter data.
[0064] For example, candidate steering control parameter data includes at least two of the following: current steering control error, vehicle speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis steering wheel. Candidate steering control parameter data may, for example, include current steering control error, vehicle speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis steering wheel.
[0065] For example, the target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed. The target steering control parameter data may include, for example, trajectory reference point curvature, current steering control error, and driving speed.
[0066] The correlation assessment information includes at least one of the correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
[0067] The correlation coefficient can be understood as a parameter that characterizes the correlation between variables.
[0068] For example, the correlation coefficient may include the Pearson correlation coefficient, which can characterize the linear correlation between variables.
[0069] The Pearson correlation coefficient can be obtained, for example, using the following formula (a):
[0070]
[0071] X and Y represent the parameter variables corresponding to two different candidate steering control parameter data.
[0072] Mutual information (MI) is used to assess and characterize the interdependence between variables. Unlike correlation coefficients, mutual information is not limited to real-valued random variables; it is more general and determines the similarity between the product of the joint distribution and the marginal distributions of the variables. Intuitively, mutual information measures the information shared by variables: it measures the degree to which knowing one of two variables reduces uncertainty about the other.
[0073] Mutual information can be obtained, for example, using the following formula (II):
[0074]
[0075] X and Y represent the parameter variables corresponding to two different candidate steering control parameter data.
[0076] For example, candidate steering parameter data with higher correlation coefficients and higher mutual information values can be selected as target steering parameter data based on numerical sorting results or filtering thresholds.
[0077] It should be noted that, unlike in straight-moving scenarios where the trajectory error of the target vehicle is related to its speed (both magnitude and direction), in turning scenarios, the trajectory error of the target vehicle is related to multiple steering parameters from the candidate steering parameters. This introduces significant uncertainty into the determination of the trajectory error data from these multiple steering parameters, resulting in inaccurate predicted trajectory error data, especially in scenarios such as high-curvature steering, lateral control, and curves.
[0078] According to the machine learning model training method of this disclosure, an initial steering control parameter sample for a target vehicle is obtained. The initial steering control parameter sample includes multiple candidate steering control parameter data related to vehicle steering. Based on the correlation evaluation information between the multiple candidate steering control parameter data, the number of target steering control parameter data is less than the number of candidate steering parameter data, and the correlation and information gain between the target steering parameter data are higher. The target steering control parameter sample includes target steering parameter data, which is more important for trajectory error prediction in steering scenarios of the target vehicle. Based on the target steering control parameter sample, a target machine learning model with more accurate trajectory error prediction adapted to scenarios such as steering, lateral control, and cornering can be trained.
[0079] Figure 3 A schematic diagram of a machine learning model training method 300 according to yet another embodiment of the present disclosure is shown.
[0080] like Figure 3 As shown, in the machine learning model training method 300 according to an embodiment of this disclosure, the initial machine learning model includes multiple branches. The N1 first learners of the first target branch of the multiple branches have different network structures. The N2 first learners of the second target branch of the multiple branches have the same network structure but different network parameters.
[0081] For example, the first learner may include at least two of the following: an eXtremeGradient Boosting (XgBoost) model, an extremely randomized Trees (ExtraTrees) model, a RandomForest model, and a GradientBoostingTree model. The second learner may include at least one of the following: a LinearRegressor model, an eXtremeGradient Boosting (XgBoost) model, and a Support Vector Regression (SVR) model.
[0082] For example, the model structure of the first learner can be more complex than that of the second learner. That is, the second learner is simpler than the first learner.
[0083] Extreme Gradient Boosting Tree (XGBoost) adds a regularization term to the objective function, significantly reducing the model's variance and simplifying the learned model, effectively preventing overfitting. XGBoost also borrows from Random Forest, supporting column sampling, which not only reduces overfitting but also decreases computation. The Gradient Boosting Tree uses the LossFunction with respect to the first derivative of f(x) to calculate pseudo-residuals for learning. XGBoost goes further, using the second derivative, which significantly improves the model's accuracy. Random Forest sampling is divided into row sampling and column sampling, where rows and columns correspond to samples and features, respectively.
[0084] For row sampling in random forests, a method with replacement is used, meaning that there may be duplicate samples in the sampled set. Assuming there are P input samples, then there are also P sampled samples. These P selected samples are used to train a decision tree, serving as the samples at the root node of the decision tree. At the same time, during training, the input samples of each tree are not all the samples, making it relatively less prone to overfitting.
[0085] For column sampling in random forests, each decision tree randomly selects q features from Q features as the node splitting features for computation. Generally, q is the square root of Q. Column sampling is further divided into two methods: global column sampling, where the same batch of sampled features is used throughout the tree construction process; and local column sampling, where m features are randomly selected separately for expansion at each node split. Column sampling further ensures that random forests do not overfit.
[0086] exist Figure 3 The example illustrates a specific instance where the initial machine learning model includes two branches, one of which is L1 and the other is L2.
[0087] One branch L1 includes four first learners 303-1 and one second learner 304-1. The four first learners include an extremely gradient boosting tree model XGB, an extremely random tree model ET, a random forest model RF, and a gradient boosting tree model GBDT. The second learner 304-1 includes a linear regression model LR.
[0088] The other branch, L2, includes K first learners 303-2 and one second learner 304-2. The K first learners consist of K extreme gradient boosting tree models XGB, from XGB-1 to XGB-K, each with different model parameters. The second learner 304-2 consists of a support vector regression model (SVR).
[0089] According to the machine learning model training method of this disclosure, the above operations enable the combination of multiple different first learners and at least one second learner to integrate multiple learners for ensemble learning. The introduction of multiple first learners introduces randomness, and the advantage of integrating multiple first learners can avoid inaccurate trajectory error predictions caused by overfitting in small samples. It can also improve the robustness and generalization of the initial machine learning model. For example, when the initial machine learning model includes multiple second learners, the inaccuracy of trajectory error predictions caused by overfitting of the first learners can be further reduced through weighted methods. The second learner learns further based on the first learner, which can alleviate the impact of overfitting.
[0090] It should be noted that, based on the correlation evaluation information between multiple candidate steering control parameter data, the number of target steering control parameter data is smaller than the number of candidate steering parameter data. This makes overfitting more likely to occur using the target steering control parameter samples used to predict trajectory error data. Furthermore, for prediction problems such as those using a target machine learning model to predict trajectory errors, as described in this embodiment, the training of the target machine learning model is more dependent on the quality of the training samples. For example, if the training samples contain significant noise or have a small amount of data, the prediction performance of the trained target machine learning model will significantly degrade. Based on these issues, the machine learning model in this embodiment can solve the overfitting problem through the above operations, resulting in a target machine learning model that can predict more accurate trajectory error data.
[0091] like Figure 3 As shown, the machine learning model training method 300 according to another embodiment of the present disclosure may further include: performing different preprocessing on the target steering control parameter samples for each branch to obtain each type of preprocessed target steering control parameter sample.
[0092] like Figure 3 As shown, the preprocessed target steering control parameter samples can, for example, be used as input to the initial machine learning model.
[0093] exist Figure 3 In the example, in the initial machine learning model, the target steering control parameter sample 301 of one branch L1 is preprocessed by standardization to obtain the target steering control parameter sample 302-1. The target steering control parameter sample 301 of the other branch L2 is preprocessed by normalization to obtain the target steering control parameter sample 302-2.
[0094] Both standardization and normalization preprocessing can transform the target steering control parameter samples. Standardization preprocessing transforms the target steering control parameter data in the sample into a distribution with a mean of 0 and a standard deviation of 1. Normalization preprocessing can scale and shift the target steering control parameter data in the sample by applying maximum and minimum values.
[0095] Different preprocessing methods result in different impacts on the training of the initial machine learning model from the preprocessed target steering control parameter samples. The target machine learning model obtained through ensemble learning can take advantage of the different preprocessing methods, resulting in a better performance and higher accuracy in predicting trajectory errors in scenarios such as steering, lateral control, and curves based on the target deep learning model.
[0096] For example, according to another embodiment of the machine learning model training method of this disclosure, a specific example of determining a feedback value based on M second output results and a target loss function can be implemented using the following embodiment: The target output result is obtained by weighting the M second output results. The feedback value is determined based on the target output result and the target loss function.
[0097] exist Figure 3 The example illustrates how weighting the two second outputs of the two second learners 304-1 and 304-2 yields the target output 305.
[0098] The target output also includes trajectory error data predicted by the current initial machine learning model.
[0099] For example, the weight of each second output result can be set to be the same or different depending on the actual scenario requirements.
[0100] According to the machine learning model training method of this disclosure, the target output result obtained by weighting M second output results can integrate the M second output results of the M second learners. The feedback value determined based on the target output result and the target loss function also comprehensively reflects the performance of the M second learners, and the feedback value is more accurate.
[0101] For example, the target loss function may include the cross-entropy loss function, etc.
[0102] This disclosure provides a method for determining autonomous driving control data, which will be described below in conjunction with... Figure 1 The system architecture, referencing Figure 4 This describes a method for determining autonomous driving control data according to exemplary embodiments of the present disclosure. The method for determining autonomous driving control data in embodiments of the present disclosure may, for example, be derived from... Figure 1 The terminal 103 shown is used to execute this.
[0103] Figure 4 A flowchart illustrating a method for determining autonomous driving control data according to an embodiment of the present disclosure is shown.
[0104] like Figure 4 As shown, the method 400 for determining autonomous driving control data in this embodiment of the present disclosure may include, for example, operations S410 to S420.
[0105] When operating S410, the target machine learning model is used to process the target steering control parameter data for the target vehicle to obtain the trajectory error data for the target vehicle.
[0106] For example, target steering control parameter data for a target vehicle can be input into a target machine learning model to obtain trajectory error data.
[0107] In operation S420, target control data for controlling the movement of the target vehicle is determined based on the trajectory error data.
[0108] For example, the target control data can correct the trajectory corresponding to the current trajectory error data.
[0109] The target machine learning model is obtained through the following operations: acquiring target steering control parameter data samples for the target vehicle and an initial machine learning model, wherein the initial machine learning model includes at least one branch, each branch including N first learners and M second learners, the N first learners are distinct, N is an integer greater than or equal to 2, and M is a positive integer less than N; inputting the target steering control parameter data samples into each first learner of the initial machine learning model to obtain N first output results; determining M second output results based on the first output results and the second learners; determining feedback values based on the M second output results and the target loss function; adjusting the model parameters of the initial machine learning model based on the feedback values to obtain the target machine learning model.
[0110] According to the method for determining autonomous driving control data in this disclosure, trajectory error prediction is performed using a target machine learning model. Since the target machine learning model's structure includes multiple branches, each branch can predict trajectory error data, and the combined data from these multiple branches provides more accurate trajectory error data. Furthermore, each branch includes N first learners and M second learners, allowing the N first learners and M second learners to learn a wider range of features related to the predicted trajectory error data, achieving ensemble learning. This results in higher accuracy in predicting trajectory error data based on target steering parameter data, adapting to scenarios such as high-curvature steering, lateral control, and curves, thus improving the driving safety and traffic efficiency of the target vehicle.
[0111] It should be noted that the target machine learning model used in the method for determining autonomous driving control data in this embodiment is trained according to the machine learning model training method of the above embodiment. The specific training process and technical effects are similar to the machine learning model training method described above, and will not be repeated here.
[0112] For example, the method for determining autonomous driving control data according to embodiments of the present disclosure further includes: acquiring initial steering control parameter data for a target vehicle, wherein the initial steering control parameter data includes multiple candidate steering control parameter data related to vehicle steering; determining target steering control parameter data based on correlation evaluation information between the multiple candidate steering control parameter data, wherein the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
[0113] For example, the candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel; the target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
[0114] Figure 5 A block diagram of an apparatus for determining autonomous driving control data according to an embodiment of the present disclosure is shown schematically.
[0115] like Figure 5 As shown, the autonomous driving control data determination device 500 of this embodiment includes, for example, a trajectory error data determination module 510 and a target control data determination module 520.
[0116] The trajectory error data determination module 510 is used to process the target steering control parameter data of the target vehicle using a target machine learning model to obtain the trajectory error data of the target vehicle.
[0117] The target control data determination module 520 is used to determine target control data for controlling the movement of the target vehicle based on the trajectory error data.
[0118] The target machine learning model is obtained using the following modules: a first acquisition module, used to acquire target steering control parameter data samples and an initial machine learning model for the target vehicle, wherein the initial machine learning model includes at least one branch, each branch includes N first learners and M second learners, the N first learners are distinct, N is an integer greater than or equal to 2, and M is a positive integer less than N; a first output result determination module, used to input the target steering control parameter data samples into each first learner of the initial machine learning model to obtain N first output results; a second output result determination module, used to determine M second output results based on the first output results and the second learners; a feedback value determination module, used to determine feedback values based on the M second output results and the target loss function; and a target machine learning model determination module, used to adjust the model parameters of the initial machine learning model based on the feedback values to obtain the target machine learning model.
[0119] For example, the device for determining autonomous driving control data according to an embodiment of the present disclosure further includes: a second acquisition module, configured to acquire initial steering control parameter data for a target vehicle, wherein the initial steering control parameter data includes multiple candidate steering control parameter data related to vehicle steering; and a target steering control parameter data determination module, configured to determine target steering control parameter data based on correlation evaluation information between the multiple candidate steering control parameter data, wherein the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
[0120] For example, in the autonomous driving control data determination device according to the embodiments of the present disclosure, the candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and chassis steering wheel maximum rate of change; and the target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
[0121] Figure 6 A block diagram of a machine learning model training apparatus according to an embodiment of the present disclosure is shown schematically.
[0122] like Figure 6 As shown, the machine learning model training device 600 of this embodiment includes, for example, a first acquisition module 610, a first output result determination module 620, a second output result determination module 630, a feedback value determination module 640, and a target machine learning model determination module 650.
[0123] The first acquisition module 610 is used to acquire target steering control parameter samples and an initial machine learning model for the target vehicle. The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners. The N first learners are all different, N is an integer greater than or equal to 2, and M is a positive integer less than N.
[0124] The first output result determination module 620 is used to input the target to the control parameter sample into each first learner of the initial machine learning model to obtain N first output results.
[0125] The second output result determination module 630 is used to determine M second output results based on the first output result and the second learner.
[0126] The feedback value determination module 640 is used to determine the feedback value based on the M second output results and the target loss function.
[0127] The target machine learning model determination module 650 is used to adjust the model parameters of the initial machine learning model based on the feedback values to obtain the target machine learning model, wherein the target machine learning model is used to determine the trajectory error data for the target vehicle.
[0128] For example, the machine learning model training apparatus according to an embodiment of the present disclosure further includes: a third acquisition module, configured to acquire an initial steering control parameter sample for a target vehicle, wherein the initial steering control parameter sample includes multiple candidate steering control parameter data related to vehicle steering; and a target steering control parameter data determination module, configured to determine target steering control parameter data based on correlation evaluation information between the multiple candidate steering control parameter data, wherein the target steering control parameter sample includes target steering control parameter data, the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
[0129] For example, in the machine learning model training apparatus according to embodiments of the present disclosure, the candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel; the target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
[0130] For example, the initial machine learning model of the machine learning model training apparatus according to the embodiments of the present disclosure includes multiple branches; the N1 first learners of the first target branch of the multiple branches have different network structures, and the N2 first learners of the second target branch of the multiple branches have the same network structure and different network parameters.
[0131] For example, the machine learning model training apparatus according to the embodiments of the present disclosure further includes: a preprocessing module, used to perform different preprocessing on the target steering control parameter samples for each branch to obtain each preprocessed target steering control parameter sample, wherein the preprocessing includes standardization preprocessing and normalization preprocessing.
[0132] For example, in the machine learning model training apparatus according to an embodiment of the present disclosure, the feedback value determination module includes: a target output result determination submodule, used to obtain a target output result by weighting M second output results; and a feedback value determination submodule, used to determine a feedback value based on the target output result and the target loss function.
[0133] For example, in the machine learning model training apparatus according to embodiments of the present disclosure, the first learner includes at least two of an extremely gradient boosting tree model, an extremely random tree model, a random forest model, and a gradient boosting tree model; the second learner includes at least one of a linear regression model, an extremely gradient boosting tree model, and a support vector regression model.
[0134] It should be understood that the embodiments of the apparatus portion of this disclosure correspond to the same or similar embodiments of the method portion of this disclosure, and the technical problems solved and the technical effects achieved are also the same or similar. This disclosure will not repeat them here.
[0135] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0136] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0137] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.
[0138] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0139] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as methods for determining autonomous driving control data and methods for training machine learning models. For example, in some embodiments, the methods for determining autonomous driving control data and methods for training machine learning models can be implemented as computer software programs, which are tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods for determining autonomous driving control data and methods for training machine learning models described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured by any other suitable means (e.g., by means of firmware) to perform methods for determining autonomous driving control data and methods for training machine learning models.
[0140] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0141] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0142] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0143] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0144] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.
[0145] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.
[0146] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0147] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A method for determining autonomous driving control data, comprising: The target vehicle's steering control parameter data is processed using a target machine learning model to obtain trajectory error data for the target vehicle. as well as Based on the trajectory error data, target control data for controlling the movement of the target vehicle is determined. The target machine learning model is obtained using the following operations: Obtain target steering control parameter data samples and an initial machine learning model for the target vehicle. The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners. The N first learners are all different, N is an integer greater than or equal to 2, and M is a positive integer less than N. The target steering control parameter data sample is input into each of the first learners of the initial machine learning model to obtain N first output results; Based on the first output result and the second learner, determine M second output results; Based on the M second output results and the target loss function, determine the feedback value; and The model parameters of the initial machine learning model are adjusted based on the feedback values to obtain the target machine learning model.
2. The method according to claim 1, further comprising: Acquire initial steering control parameter data for the target vehicle, wherein the initial steering control parameter data includes multiple candidate steering control parameter data related to vehicle steering; The target steering control parameter data is determined based on the correlation evaluation information between multiple candidate steering control parameter data, wherein the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
3. The method according to claim 2, wherein, The candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel. The target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
4. A method for training a machine learning model, comprising: Obtain target steering control parameter samples and an initial machine learning model for the target vehicle. The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners. The N first learners are all different, N is an integer greater than or equal to 2, and M is a positive integer less than N. The target steering control parameter samples are input into each of the first learners of the initial machine learning model to obtain N first output results; Based on the first output result and the second learner, determine M second output results; Based on the M second output results and the target loss function, determine the feedback value; and The model parameters of the initial machine learning model are adjusted based on the feedback values to obtain the target machine learning model, wherein the target machine learning model is used to determine the trajectory error data for the target vehicle.
5. The method according to claim 4, wherein, Also includes: Obtain an initial steering control parameter sample for the target vehicle, wherein the initial steering control parameter sample includes multiple candidate steering control parameter data related to vehicle steering; Target steering control parameter data is determined based on the correlation evaluation information between multiple candidate steering control parameter data, wherein the target steering control parameter sample includes the target steering control parameter data, the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
6. The method according to claim 5, wherein, The candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel. The target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
7. The method according to claim 4, wherein, The initial machine learning model includes multiple branches; the N1 first learners of the first target branch of the multiple branches have different network structures, and the N2 first learners of the second target branch of the multiple branches have the same network structure and different network parameters.
8. The method according to claim 7, further comprising: Different preprocessing is performed on the target steering control parameter samples for each branch to obtain each preprocessed target steering control parameter sample, wherein the preprocessing includes standardization preprocessing and normalization preprocessing.
9. The method according to any one of claims 4-8, wherein, The step of determining the feedback value based on the M second output results and the target loss function includes: The target output result is obtained by weighting the M second output results; and The feedback value is determined based on the target output and the target loss function.
10. The method according to any one of claims 4-8, wherein, The first learner includes at least two of the following: an extremely gradient boosting tree model, an extremely random tree model, a random forest model, and a gradient boosting tree model; the second learner includes at least one of the following: a linear regression model, an extremely gradient boosting tree model, and a support vector regression model.
11. A device for determining automatic driving control data, comprising: The trajectory error data determination module is used to process the target steering control parameter data of the target vehicle using a target machine learning model to obtain the trajectory error data of the target vehicle. as well as The target control data determination module is used to determine target control data for controlling the movement of the target vehicle based on the trajectory error data. The target machine learning model is obtained using the following modules: The first acquisition module is used to acquire target steering control parameter data samples and an initial machine learning model for the target vehicle. The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners. The N first learners are all different, N is an integer greater than or equal to 2, and M is a positive integer less than N. The first output result determination module is used to input the target steering control parameter data sample into each of the first learners of the initial machine learning model to obtain N first output results; The second output result determination module is used to determine M second output results based on the first output result and the second learner; The feedback value determination module is used to determine the feedback value based on M second output results and the target loss function; and The target machine learning model determination module is used to adjust the model parameters of the initial machine learning model based on the feedback values to obtain the target machine learning model.
12. The apparatus of claim 11, further comprising: The second acquisition module is used to acquire initial steering control parameter data for the target vehicle, wherein the initial steering control parameter data includes multiple candidate steering control parameter data related to vehicle steering; The target steering control parameter data determination module is used to determine the target steering control parameter data based on the correlation evaluation information between multiple candidate steering control parameter data, wherein the correlation evaluation information includes at least one of correlation coefficient and mutual information, and the number of target steering control parameter data is less than the number of candidate steering control parameter data.
13. The apparatus according to claim 12, wherein, The candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel. The target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
14. A machine learning model training device, comprising: The first acquisition module is used to acquire target steering control parameter samples and an initial machine learning model for the target vehicle. The initial machine learning model includes at least one branch, and each branch includes N first learners and M second learners. The N first learners are all different, N is an integer greater than or equal to 2, and M is a positive integer less than N. The first output result determination module is used to input the target steering control parameter sample into each of the first learners of the initial machine learning model to obtain N first output results; The second output result determination module is used to determine M second output results based on the first output result and the second learner; The feedback value determination module is used to determine the feedback value based on M second output results and the target loss function; and The target machine learning model determination module is used to adjust the model parameters of the initial machine learning model according to the feedback values to obtain the target machine learning model, wherein the target machine learning model is used to determine the trajectory error data for the target vehicle.
15. The apparatus according to claim 14, wherein, Also includes: The third acquisition module is used to acquire an initial steering control parameter sample for the target vehicle, wherein the initial steering control parameter sample includes multiple candidate steering control parameter data related to vehicle steering; The target steering control parameter data determination module is used to determine target steering control parameter data based on the correlation evaluation information between multiple candidate steering control parameter data. The target steering control parameter sample includes the target steering control parameter data, and the correlation evaluation information includes at least one of correlation coefficient and mutual information. The number of target steering control parameter data is less than the number of candidate steering control parameter data.
16. The apparatus according to claim 15, wherein, The candidate steering control parameter data includes at least two of the following: current steering control error, driving speed, acceleration, trajectory reference point curvature, vehicle maximum turning radius, and maximum rate of change of chassis and steering wheel. The target steering control parameter data includes at least two of the following: trajectory reference point curvature, current steering control error, and driving speed.
17. The apparatus according to claim 14, wherein, The initial machine learning model includes multiple branches; the N1 first learners of the first target branch of the multiple branches have different network structures, and the N2 first learners of the second target branch of the multiple branches have the same network structure and different network parameters.
18. The apparatus of claim 17, further comprising: The preprocessing module is used to perform different preprocessing on the target steering control parameter samples for each branch to obtain each preprocessed target steering control parameter sample, wherein the preprocessing includes standardization preprocessing and normalization preprocessing.
19. The apparatus according to any one of claims 14-18, wherein, The feedback value determination module includes: The target output result determination submodule is used to weight the M second output results to obtain the target output result; and The feedback value determination submodule is used to determine the feedback value based on the target output result and the target loss function.
20. The apparatus according to any one of claims 14-18, wherein, The first learner includes at least two of the following: an extremely gradient boosting tree model, an extremely random tree model, a random forest model, and a gradient boosting tree model; the second learner includes at least one of the following: a linear regression model, an extremely gradient boosting tree model, and a support vector regression model.
21. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-3 or 4-10.
22. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-3 or 4-10.
23. A computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1-3 or 4-10.
24. An autonomous vehicle, including the electronic equipment of claim 21, wherein the target vehicle drives according to the target control data.