Vehicle path tracking control method based on road information and related equipment

Abstract

The embodiment of the invention provides a vehicle path tracking control method based on road information and related equipment, and belongs to the technical field of automatic driving. The method comprises the following steps: acquiring a real-time pavement image; according to the real-time road surface image, performing image recognition through a trained road surface adhesion coefficient prediction model to obtain a road surface adhesion coefficient; performing vehicle trajectory planning according to the road adhesion coefficient to obtain trajectory data; and according to the trajectory data, path tracking is carried out through a path tracking algorithm combined with adaptive control gain so as to control vehicle driving. According to the embodiment of the invention, the adaptive capacity of the vehicle driving under different road conditions and the tracking accuracy of the vehicle to the trajectory can be improved.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a vehicle path tracking control method and related equipment based on road information. Background Technology

[0002] Intelligent vehicles have gradually become a core component of intelligent transportation systems in recent years. By integrating perception, decision-making, and control technologies, intelligent vehicles can achieve autonomous navigation in various scenarios, thereby improving road safety and optimizing traffic efficiency. However, in practical applications, the complex and ever-changing driving environment poses a severe challenge to autonomous driving systems. The coefficient of friction is a quantitative indicator of the friction between the tires and the road surface, and the differences in the coefficient of friction on different road surfaces significantly affect the vehicle's handling performance. Identifying the coefficient of friction is particularly critical. In rainy or snowy weather, with worn road surfaces or oil contamination, the coefficient of friction may drop sharply, leading to the risk of vehicle skidding or loss of control. Autonomous driving systems typically rely on preset fixed coefficients of friction or simplified empirical models for control, making it difficult to adapt to these changes in real time. For example, when a vehicle moves from a dry road surface to a slippery road section, if the control strategy is not adjusted in time, it may lead to increased braking distance or understeering, potentially causing traffic accidents. Furthermore, related technologies do not adequately utilize information about the road ahead and lack proactive adjustment capabilities, further limiting their performance in dynamic environments.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this application is to propose a vehicle path tracking control method and related equipment based on road information, which can improve the vehicle's adaptability to different road conditions and the accuracy of vehicle trajectory tracking.

[0005] To achieve the above objectives, one aspect of this application proposes a vehicle path tracking control method based on road information, the method comprising: Acquire real-time road surface images; Based on the real-time road surface image, the road surface adhesion coefficient is obtained by image recognition through a trained road surface adhesion coefficient prediction model. Vehicle trajectory planning is performed based on the road surface adhesion coefficient to obtain trajectory data; Based on the trajectory data, a path tracking algorithm incorporating adaptive control gain is used to track the path and control the vehicle's movement.

[0006] In some embodiments, the road surface adhesion coefficient prediction model includes an improved depth residual network, which is obtained through the following steps: The activation function of the original deep residual network is replaced with the activation function of the Gaussian error linear unit to obtain the improved deep residual network.

[0007] In some embodiments, the trained road surface adhesion coefficient prediction model is obtained through the following steps: Obtain a dataset of road surface images, the dataset including a training set and a validation set; The dataset is preprocessed to obtain the preprocessed training set and the preprocessed validation set; The preprocessed training set is input into the road surface adhesion coefficient prediction model, and the training results are output. The model performance metrics are calculated based on the training results and the preprocessed validation set to obtain the performance metrics. With the goal of exceeding a preset threshold for the performance index, the parameters of the road surface adhesion coefficient prediction model are updated according to a layered progressive unfreezing strategy to obtain the trained road surface adhesion coefficient prediction model.

[0008] In some embodiments, the trajectory data includes lateral acceleration and turning time on the trajectory, and the process of planning the vehicle trajectory based on the road adhesion coefficient to obtain the trajectory data includes: In response to steering commands, the vehicle's longitudinal velocity and lateral steering distance are obtained; Based on the road surface adhesion coefficient, the lateral acceleration is calculated using a preset acceleration parameter model to obtain the maximum threshold of lateral acceleration. The turning time is calculated based on the maximum threshold and the lateral distance using a preset turning trajectory function. The lateral acceleration is calculated using a preset steering trajectory function based on the longitudinal velocity and the lateral distance.

[0009] In some embodiments, the path tracking based on the trajectory data using a path tracking algorithm incorporating adaptive control gain includes: Obtain the vehicle's actual driving parameters; The steering error is obtained by calculating the error based on the trajectory data and the actual driving parameters. Based on the steering error and the actual driving parameters, the desired steering angle is obtained by processing them using a path tracking algorithm that incorporates adaptive control gain. The vehicle is steered according to the desired steering angle.

[0010] In some embodiments, the steering error includes heading angle deviation and lateral position deviation, and the path tracking algorithm is expressed as follows: ; in, This represents the desired steering angle. This indicates the deviation in the heading angle. This indicates the lateral positional deviation. This represents the adaptive control gain. For the longitudinal speed of the vehicle, This represents a preset constant.

[0011] In some embodiments, the adaptive control gain is represented as follows: ; in, Indicates the base gain. Indicates the road surface correction factor. This represents the error correction factor.

[0012] To achieve the above objectives, another aspect of this application proposes a vehicle path tracking and control device based on road information, characterized in that the device comprises: The acquisition module is used to acquire real-time road surface images; The recognition module is used to perform image recognition based on the real-time road surface image using a trained road surface adhesion coefficient prediction model to obtain the road surface adhesion coefficient. The planning module is used to plan vehicle trajectories based on the road surface adhesion coefficient to obtain trajectory data; The control module is used to perform path tracking based on the trajectory data by combining an adaptive control gain path tracking algorithm to control the vehicle's movement.

[0013] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0014] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a vehicle path tracking control method, device, electronic device, storage medium, and program product based on road information. This solution acquires real-time road surface images; based on the real-time road surface images, it performs image recognition using a trained road surface adhesion coefficient prediction model to obtain the road surface adhesion coefficient; it performs vehicle trajectory planning based on the road surface adhesion coefficient to obtain trajectory data; and based on the trajectory data, it performs path tracking using a path tracking algorithm combined with adaptive control gain to control vehicle movement. This application uses road surface images to predict the road surface adhesion coefficient in real time. The trajectory data obtained based on the road surface adhesion coefficient can reduce the risk of vehicle instability caused by unreasonable trajectories. By combining path tracking with an adaptive control gain algorithm, the vehicle can make adaptive control decisions based on the road surface adhesion coefficient, improving the accuracy of vehicle trajectory tracking, as well as the vehicle's driving performance and driving safety. Attached Figure Description

[0017] Figure 1 This is a flowchart of a vehicle path tracking control method based on road information provided in an embodiment of this application; Figure 2 This is a pseudocode diagram of the training road surface adhesion coefficient prediction model provided in the embodiments of this application; Figure 3 This is a pseudocode diagram of the trajectory planning algorithm provided in the embodiments of this application; Figure 4 This is a schematic diagram comparing lane change trajectory curves under different road conditions provided in the embodiments of this application; Figure 5 This is a schematic diagram illustrating the changes in lateral acceleration under different road surface conditions provided in the embodiments of this application; Figure 6 This is a schematic diagram illustrating the changes in vehicle heading angle under different road conditions provided in the embodiments of this application; Figure 7 This is a schematic diagram of trajectory curvature changes under different road surface conditions provided in the embodiments of this application; Figure 8 This is a schematic diagram of the Stanley geometric model structure provided in the embodiments of this application; Figure 9 This is a pseudocode diagram of the path tracking algorithm provided in the embodiments of this application; Figure 10 This is a comparison chart of the ideal trajectory and the actual trajectory under different control gains provided in the embodiments of this application; Figure 11 This is a schematic diagram illustrating the tracking error variation under different control gains provided in the embodiments of this application; Figure 12This is a schematic diagram of the lateral acceleration variation under different control gains provided in the embodiments of this application; Figure 13 This is a complete implementation flowchart of the vehicle path tracking and control method based on road information provided in the embodiments of this application; Figure 14 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Intelligent vehicles have gradually become a core component of intelligent transportation systems in recent years. By integrating perception, decision-making, and control technologies, intelligent vehicles can achieve autonomous navigation in various scenarios, thereby improving road safety and optimizing traffic efficiency. However, in practical applications, the complex and ever-changing driving environment poses a severe challenge to autonomous driving systems. The coefficient of friction is a quantitative indicator of the friction between the tires and the road surface, and the differences in the coefficient of friction on different road surfaces significantly affect the vehicle's handling performance. Identifying the coefficient of friction is particularly critical. In rainy or snowy weather, with worn road surfaces or oil contamination, the coefficient of friction may drop sharply, leading to the risk of vehicle skidding or loss of control. Traditional autonomous driving systems typically rely on preset fixed coefficients of friction or simplified empirical models for control, making it difficult to adapt to these changes in real time. For example, when a vehicle moves from a dry road surface to a slippery road section, if the control strategy is not adjusted in time, it may lead to increased braking distance or understeering, potentially causing traffic accidents. Furthermore, existing systems do not make sufficient use of information about the road ahead and lack the ability to make proactive adjustments, which further limits their performance in dynamic environments.

[0021] In the field of trajectory planning for intelligent vehicles, related technologies typically employ mathematical models based on geometric curves, such as quintic polynomials, spline curves, or Bezier curves. These methods generate smooth driving trajectories by setting the start and end points of the path and combining them with continuity constraints (such as the continuity of position, velocity, and acceleration). For example, the quintic polynomial trajectory planning method enables smooth transitions in the vehicle's motion state during lane changes or turns and is widely used in the trajectory generation modules of autonomous driving systems. However, these methods usually assume a fixed road adhesion coefficient or are based on empirical parameters, failing to adequately consider the dynamic changes in actual road conditions.

[0022] Related technologies employ fixed kinematic parameters (such as maximum lateral acceleration) and adhesion coefficient assumptions during trajectory planning, resulting in trajectories that fail to adapt to changes in road conditions in real time. Specifically, this method lacks a real-time sensing and feedback mechanism for the road adhesion coefficient, leading to planned trajectories that may exceed the vehicle's actual grip limit on low-adhesion surfaces (such as wet or icy roads), increasing the risk of skidding or loss of control. Furthermore, these methods lack utilization of forward-looking road information and fail to proactively adjust path parameters during trajectory generation, thus limiting their safety and adaptability in dynamic environments.

[0023] In view of this, this application provides a vehicle path tracking control method and related equipment based on road information. This method uses road surface images to estimate the road surface adhesion coefficient in real time. The trajectory data obtained based on the road surface adhesion coefficient can reduce the risk of vehicle instability caused by unreasonable trajectory. By combining a path tracking algorithm with adaptive control gain, the vehicle can make adaptive control decisions based on the road surface adhesion coefficient, thereby improving the accuracy of the vehicle's trajectory tracking, as well as the vehicle's driving performance and driving safety.

[0024] This application provides a vehicle path tracking and control method based on road information, relating to the field of autonomous driving technology. The vehicle path tracking and control method based on road information provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the vehicle path tracking and control method based on road information, but is not limited to the above forms.

[0025] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0026] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0027] Figure 1This is an optional flowchart of the vehicle path tracking control method based on road information provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0028] Step S101: Obtain real-time road surface image; Step S102: Based on the real-time road surface image, perform image recognition using the trained road surface adhesion coefficient prediction model to obtain the road surface adhesion coefficient. Step S103: Perform vehicle trajectory planning based on the road surface adhesion coefficient to obtain trajectory data; Step S104: Based on the trajectory data, path tracking is performed using a path tracking algorithm that incorporates adaptive control gain to control vehicle movement.

[0029] Steps S101 to S104 as shown in the embodiments of this application estimate the road surface adhesion coefficient in real time through road surface images. The trajectory data obtained based on the road surface adhesion coefficient can reduce the risk of vehicle instability caused by unreasonable trajectory. By combining the path tracking algorithm with adaptive control gain, the vehicle can make adaptive control decisions based on the road surface adhesion coefficient, thereby improving the accuracy of the vehicle's trajectory tracking, as well as the vehicle's driving performance and driving safety.

[0030] In step S101 of some embodiments, the image acquisition device can be mounted on the windshield of the vehicle to acquire real-time road surface images. The real-time road surface images include binocular RGB images and depth information, serving as the basis for subsequent road surface analysis and estimation of the road surface adhesion coefficient.

[0031] In step S102 of some embodiments, the road surface adhesion coefficient prediction model is a deep learning model. The trained deep learning model is used to analyze the road surface type of the real-time road surface image and predict the real-time road surface adhesion coefficient. The road surface adhesion coefficient serves as the basis for subsequent vehicle path planning and control decisions.

[0032] In some embodiments, the road surface adhesion coefficient prediction model can be an improved deep residual network, which can be obtained through the following step S210: Step S210: Replace the activation function of the original deep residual network with the Gaussian error linear unit activation function to obtain the improved deep residual network.

[0033] In step S210 of some embodiments, the deep residual network can be ResNet50. Since the fully connected layer structure of ResNet50 is simple, it is easy to lose high-order features. The original ReLU activation function of ResNet50 has a gradient discontinuity at the zero point. This mathematical characteristic may restrict the optimization process of the model. In addition, when a deterministic activation mechanism is used in deep architectures such as ResNet50, a random regularization mechanism (such as Dropout) is usually required to enhance the generalization ability of the model. Therefore, the activation function of the improved deep residual network adopts the Gaussian Error Linear Unit (GELU) activation function. The GELU activation function relies on the probability statistics of the input itself to provide random regularization for the activation function, achieving implicit regularization while protecting the integrity of feature information, as shown in Equation (1).

[0034] (1) in, It refers to x The cumulative function of the Gaussian normal distribution is expressed as shown in equation (2).

[0035] (2) Compared to ReLU's hard-threshold activation mechanism, GELU, by introducing a probabilistic modeling method, preserves the asymptotic nonlinear response characteristics in the negative region. Its continuously differentiable mathematical properties not only effectively alleviate the vanishing gradient phenomenon in neurons but also improve the convergence stability of the function in the zero-neighborhood through smooth gradient transitions. GELU's probabilistic activation characteristics are more suitable for handling the ambiguous boundary problems on icy and snowy roads.

[0036] In some embodiments, the trained road surface adhesion coefficient prediction model can be obtained through steps S201 to S205: Step S201: Obtain a dataset of road surface images, which includes a training set and a validation set; Step S202: Preprocess the dataset to obtain a preprocessed training set and a preprocessed validation set; Step S203: Input the preprocessed training set into the road surface adhesion coefficient prediction model and output the training results; Step S204: Calculate the model performance metrics based on the training results and the preprocessed validation set to obtain the performance metrics; Step S205: With the goal of exceeding the preset threshold for performance indicators, the parameters of the road surface adhesion coefficient prediction model are updated according to the layered progressive unfreezing strategy to obtain the trained road surface adhesion coefficient prediction model.

[0037] In step S201 of some embodiments, the dataset includes a public dataset and real-vehicle acquired data. The images in the dataset include four road surface types: snow and ice, wet, dry, and unpaved dirt roads. The snow and ice and wet road surface images in the public dataset are from the DAWN dataset. The real-vehicle dataset consists of two types of images: dry asphalt and unpaved dirt roads, acquired by a camera, which are combined with the wet and snow and ice and ice road surface datasets to create a mixed dataset. Based on stereo vision perception technology, a ZED 2i binocular camera was used to acquire images of dry asphalt roads (including straight roads, curves, intersections, etc.) and unpaved dirt roads in an urban area. The acquisition device was mounted on the windshield of the experimental vehicle. The binocular RGB images and depth information were acquired simultaneously at a resolution of 2560×720 and a frame rate of 30fps using the SDK tool. The specific optical parameters and performance indicators of the ZED 2i camera used for data acquisition are shown in Table 1.

[0038] Table 1 ZED 2i Camera Parameters

[0039] After processing, a total of 97 dry asphalt pavement images and 40 unpaved dirt road images were collected from the real vehicle data. The public dataset and the real vehicle data were mixed as the training set and the test set. There are four types of road surfaces: snow and ice, wet, dry, and unpaved dirt roads, with 204, 200, 97, and 40 images respectively, for a total of 541 images. The ratio of training set to validation set is 8:2.

[0040] In step S202 of some embodiments, a data processing module preprocesses the dataset. For the training set, random geometric transformations (random cropping to 224×224, horizontal flip probability 0.5, rotation ±20°) and color space perturbations (brightness / contrast jitter amplitude 0.3, random erasure probability 0.2) are applied. For the validation set, deterministic center cropping and standardization (ImageNet mean-variance normalization) are retained. This embodiment improves data diversity while maintaining the stability of the validation set evaluation.

[0041] In steps S203 to S205 of some embodiments, performance metrics and loss values ​​are calculated based on the training results and the preprocessed validation set. The performance metrics include accuracy, precision, recall, and F1 score, and the loss value is calculated using a loss function.

[0042] Traditional transfer learning methods typically employ a global freeze strategy, leading to insufficient adaptability of deep features. In this embodiment, a layered progressive unfreezing strategy is used during training: the first three residual blocks of ResNet50 (layers 1-3) are frozen, while layer 4 and the fully connected layer are unfrozen. Freezing fixes the parameters of layers 1-3, preventing updates during training, while unfreezing allows the parameters of layer 4 and the fully connected layer to be updated. This design is based on feature hierarchy theory, where shallower networks primarily extract general edge texture features, while deeper networks learn domain-specific features. A smooth transition in feature extraction is achieved through layered learning rate settings (1e-4 → 1e-3). This layered progressive unfreezing strategy enhances the adaptability of higher-level features while maintaining the stability of basic features.

[0043] Furthermore, during the training process, a learning rate scheduler is used to adjust the model's learning rate. The learning rate scheduler, ReduceLROnPlateau, adjusts the learning rate based on the F1 score on the validation set. When the validation F1 score does not improve for several consecutive epochs, the learning rate is halved. This helps the model to fine-tune parameters more effectively in later stages and reduces oscillations. In addition, class weights are added to the loss function to address class imbalance and reduce the likelihood of the model being biased towards the majority class.

[0044] By adjusting model parameters to improve performance metrics and reduce loss values, and saving the current model parameters when the performance metrics exceed a preset threshold, a well-trained road adhesion coefficient prediction model can be obtained. The pseudocode for the training process is as follows: Figure 2 As shown.

[0045] This embodiment uses core metrics such as accuracy, precision, recall, and F1 score to jointly evaluate classification performance, numerically representing the model's discriminative ability and predictive reliability. Accuracy refers to the proportion of correctly predicted positive and negative examples out of the total number of examples; precision refers to the proportion of correctly predicted positive examples out of the total number of actual positive examples; recall refers to the proportion of correctly predicted positive examples out of the total number of actual positive examples, primarily assessing the coverage of whether all actual positive examples were predicted; the F1 score is the harmonic mean of precision and recall, and its high harmonic mean is only achieved when both precision and recall are at high levels.

[0046] Table 2 shows the performance comparison of the three models, indicating that the improved ResNet50 model performs best. Based on the evaluation metrics, the improved ResNet50 model architecture, through multi-level optimization, significantly improves road classification performance, achieving an accuracy of 99.17% and an F1 score of 99.17%. This is a significant improvement over the 95.87% accuracy of the ResNet50 model using the global freezing strategy in traditional transfer learning methods, and also surpasses the 96.69% accuracy of the classic deep convolutional neural network model VGG16.

[0047] Table 2 Performance Comparison Results of the Three Models

[0048] Based on GA / T 643—2006 and automotive theory, the corresponding ranges of road surface types and road surface adhesion coefficients are obtained, as shown in Table 3.

[0049] Table 3 Comparison of Road Surface Adhesion Coefficients

[0050] In step S103 of some embodiments, the ideal lane-changing trajectory should meet multi-dimensional performance indicators. First, in terms of geometric characteristics, the trajectory curve should remain continuous and differentiable with a smooth curvature transition, while minimizing the longitudinal travel distance to reduce the impact on traffic flow in adjacent lanes. The curvature amplitude of the trajectory should be controlled within a reasonable range, and the curvature and its first derivative at the start and end points of the trajectory should both be zero, so that the front wheel steering angle of the vehicle is zero at the initial and completion moments of the lane change, maintaining lateral handling stability.

[0051] At the kinematic constraint level, it is necessary to limit the rate of change of the vehicle's lateral acceleration (i.e., jerk) to improve ride comfort. The lateral displacement at the end of the trajectory should precisely match the standard lane width, and the heading angles and their rates of change at the start and end points should be zero, so that the vehicle always maintains a parallel relationship with the lane lines before and after lane changes, reducing unnecessary positional deviations.

[0052] By combining the estimated adhesion coefficient, vehicle state, and environmental constraints, a smooth and safe trajectory is generated using a constant velocity offset model and a sine function, forming a trajectory planning based on dynamic constraints of the road adhesion coefficient, and obtaining relevant trajectory data.

[0053] In some embodiments, step S103 may include, but is not limited to, steps S301 to S304: Step S301: In response to the steering command, obtain the vehicle's longitudinal speed and the lateral distance of the steering. Step S302: Based on the road surface adhesion coefficient, calculate the lateral acceleration using a preset acceleration parameter model to obtain the maximum threshold of the lateral acceleration. Step S303: Based on the maximum threshold and lateral distance, the steering time is calculated using a preset steering trajectory function. Step S304: Based on the longitudinal velocity and lateral distance, the lateral acceleration is calculated using a preset steering trajectory function.

[0054] In step S301 of some embodiments, the trajectory data includes lateral displacement, longitudinal displacement, lateral acceleration and trajectory curvature at multiple times. Taking vehicle lane change as an example, the trajectory of vehicle lane change is a function based on the superposition of sine function and constant velocity offset model, as shown in equation (3).

[0055] (3) By weighting and superimposing the parameters of equation (3), the lane change trajectory function (steering trajectory function) is expressed as shown in equation (4).

[0056] (4) in Lateral displacement (lateral distance) for the vehicle to complete the entire lane-changing process. This refers to the longitudinal displacement of the vehicle during the lane-changing process. The longitudinal speed of the vehicle is represented by equation (4), and the target turning distance can be expressed by equation (4). During lane changing, the longitudinal speed is set to a constant value, while... The time required to complete the entire lane change process can be obtained from equation (5).

[0057] (5) Substituting equation (5) into equation (4) and taking the second derivative, we can obtain the relationship between the lateral acceleration of the vehicle and time during the lane change process, as shown in equation (6).

[0058] (6) in, This refers to the lateral acceleration of the vehicle during lane changing.

[0059] In step S302 of some embodiments, the acceleration parameter model is shown in equation (7).

[0060] (7) in, For the longitudinal speed of the vehicle, g It is the acceleration due to gravity. Let be the road surface adhesion coefficient. The maximum threshold for lateral acceleration can be obtained according to equation (7). .

[0061] From the above formulas (5) and (6), the time when the maximum value of lateral acceleration occurs can be obtained. tAs shown in equation (8).

[0062] (8) By differentiating the lateral acceleration and setting the derivative to zero, we can obtain the lane change time (steering duration). As shown in equation (9).

[0063] (9) In an embodiment of the lane change trajectory generation method, the algorithm can output feasible paths under different combinations of road surface adhesion coefficients and longitudinal speeds. The pseudocode of the trajectory planning algorithm is as follows: Figure 3 As shown, Figure 3 The calculation process for multiple parameters of the trajectory data is demonstrated.

[0064] To verify the effectiveness of the trajectory planning in the above embodiments, related experiments were conducted. In the experimental configuration, the vehicle's longitudinal speed was set to a constant 54 km / h, and lane changes were set to be completed in adjacent lanes. A standard lane width of 3.75 m was used. The lane change time was derived from the constraint of the maximum lateral acceleration. Referring to Table 3, which identifies the range of road adhesion coefficients for the three types of road surfaces, this experiment used dry asphalt pavement with μ=0.75, wet asphalt pavement with μ=0.45, and snow-covered asphalt pavement with μ=0.2. The resulting lane change trajectory curves are shown below. Figure 4 As shown, under simulation scenarios with varying road surface adhesion conditions, lane-changing trajectory curves for three typical road conditions—dry, wet, and icy / snowy—wet and icy, were generated based on the lateral dynamic constraint thresholds in the preset parameter table. The data indicates that as the road surface adhesion coefficient decreases, the vehicle lane-changing time increases, the trajectory curvature is smaller and smoother, effectively improving ride comfort.

[0065] Lateral acceleration during lane changes under three road surface adhesion conditions, as follows Figure 5 As shown. By Figure 5 Therefore, when the road surface conditions are good, such as a dry asphalt road surface with μ=0.75, the lateral acceleration amplitude is 3.311 m / s². 2 When road conditions are poor, such as wet asphalt pavement μ=0.45 and snow-covered asphalt pavement μ=0.2, the lateral acceleration amplitude is 1.545 m / s². 2 and 1.099m / s 2 When road conditions are poor, trajectory planning can be performed based on the identification results of the road adhesion coefficient, reducing lateral acceleration and sharp steering, resulting in better stability and comfort.

[0066] Figure 6 The vehicle heading angle under different road surface conditions changes smoothly without drastic fluctuations. Figure 7 The curves of the trajectory under different road surface conditions are shown in the figure. It can be seen from the figure that the curvature of the lane change trajectory is continuous and smooth.

[0067] Table 4 shows the lane change time, maximum lateral acceleration, maximum heading angle, and curvature amplitude under different road surface conditions.

[0068] Table 4 Evaluation Indicators for Vehicle Lane Changing under Different Road Surface Conditions

[0069] In step S104 of some embodiments, in the field of path tracking control of autonomous driving systems, geometric path tracking algorithms, as a typical non-model-driven control strategy, achieve trajectory tracking by constructing a geometric mapping relationship between the vehicle pose and the reference path. The core feature of this algorithm is the use of a forward-looking error observation mechanism, which establishes an error compensation strategy based on spatial curvature by setting virtual preview points. Specifically, the system generates a feedback control law with geometric constraints by real-time calculation of the lateral deviation and heading angle deviation of the preview distance projection point. This type of algorithm has the advantages of low complexity and strong real-time performance.

[0070] In some embodiments, step S104 may include, but is not limited to, steps S401 to S404: Step S401: Obtain the actual driving parameters of the vehicle; Step S402: Calculate the steering error based on the trajectory data and actual driving parameters; Step S403: Based on the steering error and actual driving parameters, the desired steering angle is obtained by processing the path tracking algorithm combined with adaptive control gain. Step S404: Control the vehicle to steer according to the desired steering angle.

[0071] In steps S401 and S404 of some embodiments, the actual driving parameters may include the vehicle's speed, heading angle, and real-time lateral displacement. Steering error includes heading angle error and lateral position deviation. The path tracking algorithm may be the Stanley algorithm, a nonlinear feedback function based on lateral tracking error that achieves exponential convergence of the control error. The Stanley method adjusts the front wheel steering angle to align the vehicle's front wheel direction with the desired path tangent direction, while compensating for lateral errors. The Stanley geometric model is as follows: Figure 8 As shown.

[0072] like Figure 2 As shown, heading angle error As shown in equation (10).

[0073] (10) in, For the vehicle's heading angle, and Reference path point The tangent angle.

[0074] Compared to the traditional Pure Pursuit tracking method, the Stanley controller integrates the coupling relationship between heading and position deviation to form a regulation mechanism with feedforward-feedback composite characteristics, which improves path tracking accuracy while significantly reducing sensitivity to sudden changes in path curvature.

[0075] Stanley's core control formula combines heading angle deviation and lateral position deviation. Based on this, an adaptive gain mechanism (adaptive control gain) is designed. The path tracking algorithm combined with the adaptive control gain is shown in Equation (11).

[0076] (11) in, Indicates the desired steering angle. Indicates the deviation in heading angle. Indicates lateral positional deviation. Indicates the adaptive control gain. For the longitudinal speed of the vehicle, This represents a preset constant. It can also be a zero constant (default is 0.01m / s) to prevent the denominator from approaching zero at low speeds.

[0077] The adaptive control gain is shown in equation (12).

[0078] (12) in, Indicates the base gain. Indicates the road surface correction factor. This represents the error correction factor.

[0079] This is achieved through a pre-calibrated lookup table or piecewise linear function; the higher the road surface adhesion coefficient μ, the better. The closer the coefficient of friction μ is to 1, the higher the responsiveness of the control system; the lower the coefficient of friction μ, the better. The smaller the value, the more actively the control gain is reduced, ensuring operational stability. A specific implementation is shown in equation (13).

[0080] (13) in, This is the upper limit threshold for low adhesion coefficients, such as for icy and snowy roads; This is the lower threshold for high adhesion coefficients, corresponding to dry asphalt pavement; The minimum gain coefficient to ensure road safety.

[0081] The control gain is dynamically fine-tuned based on the error of path tracking. The purpose is to prevent overshoot and oscillation when the error is large, reduce the gain when the initial error is large to avoid lateral acceleration exceeding the limit, and increase the gain when the error is small to improve tracking accuracy. A specific implementation method is to use the hyperbolic tangent function to avoid the gain change being too drastic near the zero point of the error, while ensuring sufficient attenuation effect when the error is large, as shown in equation (14).

[0082] (14) in, The attenuation coefficient ( ), used to control the rate at which the gain decays as the error increases, This is the minimum value of the error correction coefficient.

[0083] The pseudocode for the path tracing algorithm is as follows: Figure 9 As shown.

[0084] To evaluate the performance of the path tracking algorithm with adaptive control gain described in the above embodiments, the Stanley controller module introduces an actuator dynamic constraint model based on the classical control law, limiting the steering angle amplitude to ±30°, effectively simulating the inertial characteristics of a real steering system. The controller's error compensation mechanism considers both heading angle deviation and lateral position deviation. Multi-dimensional performance evaluation is based on three quantitative indicators: overshoot, root mean square error, and steering angle.

[0085] In the evaluation index section, the RMSE (Root Mean Square Error) used reflects the average deviation between the actual trajectory of the vehicle and the reference trajectory. Its mathematical definition is shown in Equation (15).

[0086] (15) Where n is the number of sampling points, Let i be the lateral position of the reference trajectory at the i-th sampling point. Let be the longitudinal position of the reference trajectory at the i-th sampling point.

[0087] Overshoot represents the maximum percentage by which a vehicle exceeds the target position during tracking, as defined in equation (16).

[0088] (16) in, This represents the lateral position where the lateral deviation between the actual trajectory and the reference trajectory is at its maximum. This represents the lateral position of the reference trajectory when the corresponding lateral deviation is at its maximum.

[0089] In the implementation of the path tracking algorithm, the vehicle is also set to move at a constant longitudinal speed of 54 km / h. Lane changes are set to be completed in the adjacent lane. The standard lane width is 3.75m, the wet asphalt road surface μ=0.45 is selected, the vehicle wheelbase is 2.7m, the initial position is selected as x=0, y=0, and the initial heading angle is set to 0. The velocity is 0.01 m / s, and the control gain K is selected as 0.5, 1, 3, and 5 respectively. The ideal trajectory and actual trajectory curves for different K values ​​are shown below. Figure 10 As shown, the tracking error under different K values ​​is as follows: Figure 11 As shown. In vehicle path trajectory tracking control, by Figure 10 and Figure 11 When K = 0.5, 1, 3, and 5, the overshoot decreases monotonically with K by 4.4%, 4.3%, 1.7%, and 0.8%, respectively, with RMSE values ​​of 0.111m, 0.098m, 0.058m, and 0.042m. In the Stanley control algorithm, heading correction dominates the initial stage of the transient response, affecting the heading error. Dominant steering angle This allows for rapid adjustment of vehicle orientation, reducing accumulated lateral error. Lateral correction dominates steady-state accuracy; a high K value is sensitive to the lateral error e through the arctan term, driving... Lateral acceleration at different K values, such as Figure 12 As shown.

[0090] Table 5 shows the impact of control gain K on vehicle dynamic parameters (jump, maximum lateral acceleration) and tracking accuracy evaluation indicators (overshoot, RMSE). The data in Table 5 show that at low gain, overshoot and error are higher, but jump and acceleration are relatively lower. At K=3, jump and acceleration reach their peak values, but overshoot and error further decrease. At K=5, both overshoot and RMSE are lowest, and jump and acceleration are slightly improved compared to K=3.

[0091] Table 5. Effect of Control Gain K

[0092] In some embodiments, please refer to Figure 13 .

[0093] (1) Design a road surface adhesion coefficient prediction method based on deep residual networks. A layered progressive unfreezing strategy is adopted, which enhances the adaptability of high-level features while maintaining the stability of basic features, making it suitable for road surface recognition tasks.

[0094] (2) Constructing Gaussian error linear units. By introducing probabilistic modeling methods, Gaussian error linear units retain asymptotic nonlinear response characteristics in the negative region. Its continuously differentiable mathematical properties not only effectively alleviate the vanishing gradient phenomenon of neurons, but also improve the convergence stability of the function in the neighborhood of zeros through smooth gradient transition. The probabilistic activation characteristics of GELU are more suitable for handling the fuzzy boundary problem of icy and snowy roads.

[0095] (3) Add a learning rate scheduler and a data processing module. In the training process, the learning rate scheduler ReduceLROnPlateau adjusts the learning rate based on the F1 score of the validation set. In the data processing module, the training set adopts random geometric transformation and color space perturbation, while the validation set retains deterministic center pruning and standardization operations, which improves data diversity while maintaining the stability of the validation set evaluation.

[0096] (4) Design the constraints that the ideal lane-changing trajectory must satisfy. The lane-changing trajectory planning algorithm needs to balance computational efficiency and engineering practicality. Under the premise of satisfying the constraints described in the previous embodiments, a low-complexity mathematical model is constructed, and an efficient solution strategy is adopted to enhance the real-time response capability of the system, so that a safe and feasible trajectory scheme can be generated quickly in a dynamic traffic environment. This multi-constraint collaborative optimization method can not only effectively avoid obstacles, but also achieve a comprehensive balance between driving smoothness, handling stability and computational economy.

[0097] (5) Design the intelligent vehicle lane-changing trajectory based on the superposition of sine function and constant velocity offset model. The function based on the superposition of sine function and constant velocity offset model is selected as the trajectory of intelligent vehicle lane changing.

[0098] (6) The road surface type obtained by the image recognition module is used to estimate the road surface adhesion coefficient. Based on this, the lateral acceleration is limited, and then the lane change time is calculated to obtain the lane change trajectory.

[0099] (7) Calculate the steering control amount based on the path tracking deviation of the vehicle's front axle center point. The Stanley algorithm is a nonlinear feedback function based on lateral tracking error, which can achieve exponential convergence of control error. The Stanley method adjusts the front wheel steering angle to make the direction of the vehicle's front wheels consistent with the tangential direction of the desired path, while compensating for lateral error.

[0100] (8) Design a gain adaptive mechanism. Stanley control combines heading deviation and lateral position deviation, and then designs a gain adaptive mechanism based on this. When the initial error is large, the gain is reduced to avoid lateral acceleration exceeding the limit, and when the error is small, the gain is increased to improve tracking accuracy.

[0101] (9) Construct a complete simulation process integrating trajectory generation, controller design, and performance evaluation. The path tracking module, based on the Stanley algorithm, constructs a complete simulation process integrating trajectory generation, controller design, and performance evaluation, suitable for path tracking control research of autonomous vehicles. As shown above, the trajectory generation module, based on the classical control law, introduces an actuator dynamic constraint model and a steering angle amplitude limit to effectively simulate the inertial characteristics of the actual steering system. The controller's error compensation mechanism considers both heading angle deviation and lateral position deviation. The multi-dimensional performance evaluation module proposes three quantitative indicators: overshoot, root mean square error, and steering angle.

[0102] The embodiments of this application have at least the following beneficial effects: In terms of road surface adhesion coefficient estimation, by introducing a hierarchical progressive thawing strategy and the GELU activation function, the improved ResNet50 model can more accurately identify different road surface types and predict adhesion coefficients. This model can dynamically adjust feature extraction weights based on subtle differences in road surface characteristics and enhance its ability to identify ambiguous boundaries (such as transition areas between icy and wet road surfaces) through a probabilistic activation mechanism, significantly improving its generalization performance and enabling it to more accurately adapt to diverse road scenarios. This not only helps in obtaining crucial road surface state information in real time but also provides a reliable decision-making basis for subsequent trajectory planning and path tracking for intelligent vehicles, enhancing their environmental perception capabilities in complex road conditions.

[0103] In constructing the trajectory planning model, this embodiment establishes a lane-change trajectory planning method based on dynamic constraints of the road surface adhesion coefficient. A continuous and smooth trajectory is generated by superimposing a sine function and a constant-velocity offset model, improving the continuity of vehicle kinematic parameters (lateral acceleration, curvature, etc.). The lateral acceleration threshold is adjusted in real time based on the estimated road surface adhesion coefficient, ensuring that the trajectory meets safety and comfort constraints under different adhesion conditions. The trajectory model, established by combining kinematic characteristics and road surface conditions, comprehensively considers factors such as safety, comfort, and efficiency, providing support for generating the optimal driving path and helping to reduce the risk of vehicle instability caused by unreasonable trajectories, thus improving the engineering practicality and environmental adaptability of the trajectory.

[0104] In terms of path tracking control optimization, the Stanley algorithm dynamic gain adjustment mechanism proposed in this embodiment enables intelligent vehicles to make adaptive control decisions based on the road adhesion coefficient. During tracking, the vehicle calculates the tracking performance (overshoot, root mean square error, etc.) under different gains in real time, achieving the goal of minimizing the risk of lateral acceleration exceeding limits while ensuring tracking accuracy. In low-adhesion road surfaces or scenarios with large initial errors, this mechanism can automatically reduce the control gain to prevent the vehicle from sideslipping due to oversteering; in high-adhesion road surfaces and scenarios with small errors, the gain is increased to improve tracking accuracy, achieving a balance in control performance under different scenarios.

[0105] This application embodiment also provides a vehicle path tracking and control device based on road information, which can implement the above method. The device includes: The acquisition module is used to acquire real-time road surface images; The recognition module is used to perform image recognition based on real-time road surface images using a trained road surface adhesion coefficient prediction model to obtain the road surface adhesion coefficient. The planning module is used to plan vehicle trajectories based on the road surface adhesion coefficient and obtain trajectory data. The control module is used to track the path based on trajectory data using a path tracking algorithm that incorporates adaptive control gain, in order to control the vehicle's movement.

[0106] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0107] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0108] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0109] Please see Figure 14 , Figure 14 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the methods described in the embodiments of this application. The 903 input / output interface is used to implement information input and output. The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0110] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0111] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0112] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0113] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0114] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0115] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0116] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0117] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0118] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0119] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0120] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0122] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A vehicle path tracking control method based on road information, characterized in that, The method includes the following steps: Acquire real-time road surface images; Based on the real-time road surface image, the road surface adhesion coefficient is obtained by image recognition through a trained road surface adhesion coefficient prediction model. Vehicle trajectory planning is performed based on the road surface adhesion coefficient to obtain trajectory data; Based on the trajectory data, a path tracking algorithm incorporating adaptive control gain is used to track the path and control the vehicle's movement.

2. The method according to claim 1, characterized in that, The road surface adhesion coefficient prediction model includes an improved depth residual network, which is obtained through the following steps: The activation function of the original deep residual network is replaced with the activation function of the Gaussian error linear unit to obtain the improved deep residual network.

3. The method according to claim 1, characterized in that, The trained road surface adhesion coefficient prediction model is obtained through the following steps: Obtain a dataset of road surface images, the dataset including a training set and a validation set; The dataset is preprocessed to obtain the preprocessed training set and the preprocessed validation set; The preprocessed training set is input into the road surface adhesion coefficient prediction model, and the training results are output. The model performance metrics are calculated based on the training results and the preprocessed validation set to obtain the performance metrics. With the goal of exceeding a preset threshold for the performance index, the parameters of the road surface adhesion coefficient prediction model are updated according to a layered progressive unfreezing strategy to obtain the trained road surface adhesion coefficient prediction model.

4. The method according to claim 1, characterized in that, The trajectory data includes lateral acceleration and turning time on the trajectory. The process of planning the vehicle trajectory based on the road surface adhesion coefficient to obtain trajectory data includes: In response to steering commands, the vehicle's longitudinal velocity and lateral steering distance are obtained; Based on the road surface adhesion coefficient, the lateral acceleration is calculated using a preset acceleration parameter model to obtain the maximum threshold of lateral acceleration. The turning time is calculated based on the maximum threshold and the lateral distance using a preset turning trajectory function. The lateral acceleration is calculated using a preset steering trajectory function based on the longitudinal velocity and the lateral distance.

5. The method according to claim 1, characterized in that, The path tracking based on the trajectory data, using a path tracking algorithm incorporating adaptive control gain, includes: Obtain the vehicle's actual driving parameters; The steering error is obtained by calculating the error based on the trajectory data and the actual driving parameters. Based on the steering error and the actual driving parameters, the desired steering angle is obtained by processing them using a path tracking algorithm that incorporates adaptive control gain. The vehicle is steered according to the desired steering angle.

6. The method according to claim 5, characterized in that, The steering error includes heading angle deviation and lateral position deviation, and the path tracking algorithm is expressed as follows: ； in, This represents the desired steering angle. This indicates the deviation in the heading angle. This indicates the lateral positional deviation. This represents the adaptive control gain. For the longitudinal speed of the vehicle, This represents a preset constant.

7. The method according to claim 6, characterized in that, The adaptive control gain is expressed as follows: ； in, Indicates the base gain. Indicates the road surface correction factor. This represents the error correction factor.

8. A vehicle path tracking and control device based on road information, characterized in that, The device includes: The acquisition module is used to acquire real-time road surface images; The recognition module is used to perform image recognition based on the real-time road surface image using a trained road surface adhesion coefficient prediction model to obtain the road surface adhesion coefficient. The planning module is used to plan vehicle trajectories based on the road surface adhesion coefficient to obtain trajectory data; The control module is used to control the vehicle's movement by performing path tracking based on the trajectory data using a path tracking algorithm that incorporates adaptive control gain.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images