Automatic driving automobile dynamic trajectory planning method based on transverse and longitudinal coordination

Abstract

The invention relates to the technical field of automatic driving track planning, and discloses an automatic driving automobile dynamic track planning method based on transverse and longitudinal coordination. The method comprises the following steps: acquiring transverse motion state information and longitudinal motion state information of the self-driving automobile, and acquiring dynamic information of a surrounding environment; according to the collected transverse and longitudinal motion state information, calculating to obtain a transverse and longitudinal coordination degree index; inputting the dynamic information of the surrounding environment into a preset dynamic trajectory model, and outputting a reference transverse trajectory parameter and a reference longitudinal trajectory parameter; determining a trajectory planning deviation value by analyzing the deviation between the transverse and longitudinal coordination degree indexes and the reference trajectory parameters; and correcting a transverse control parameter and a longitudinal control parameter in the track generation process based on the deviation value. According to the method, through transverse and longitudinal coordination analysis and dynamic deviation correction, the planned track better fits vehicle motion characteristics and environment changes, and the rationality and dynamic adaptability of the automatic driving track are improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving trajectory planning technology, specifically to a dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination. Background Technology

[0002] In the development of autonomous driving technology, trajectory planning, as a core link connecting environmental perception and vehicle control, directly affects the safety, smoothness, and efficiency of vehicle operation. With the increasing complexity of scenarios such as urban roads and highways, and the dynamic changes of pedestrians and other traffic participants, higher demands are placed on the dynamic adaptability and multi-dimensional coordination of autonomous driving trajectory planning.

[0003] Currently, most autonomous driving trajectory planning methods employ a separate design approach when handling lateral and longitudinal motion. Some methods prioritize the generation of the lateral trajectory, such as path smoothing based on reference lines, and then match the longitudinal speed planning based on the lateral trajectory results. Other methods focus on longitudinal speed control, adjusting the driving speed through preset acceleration and deceleration thresholds, with the lateral path serving only as an auxiliary reference. This separate approach easily overlooks the inherent relationship between lateral motion (such as steering angle and lateral acceleration) and longitudinal motion (such as vehicle speed and longitudinal acceleration). For example, in emergency avoidance scenarios, if lateral steering adjustments have caused the vehicle's lateral acceleration to approach a safe threshold, while the longitudinal speed is still maintained at a high level as originally planned, it may lead to vehicle instability, increasing the risk of rollover or collision. In cornering scenarios, if steering is planned only based on the lateral curvature of the curve without simultaneous coordination of longitudinal speed, the cornering speed may be too high or too low, affecting driving efficiency and ride comfort.

[0004] Existing trajectory planning methods have limitations in utilizing dynamic information about the surrounding environment. While some methods collect environmental information such as obstacle positions and traffic light status, they only incorporate this information as static constraints into trajectory generation, failing to capture dynamic changes in environmental information in real time. For example, if a vehicle ahead suddenly decelerates, and the trajectory model still generates a reference trajectory based on historical speed data, the reference parameters are likely to deviate significantly from the actual environment, leading to delayed adjustments in subsequent control parameters and an inability to respond promptly to emergencies. Furthermore, in the trajectory deviation correction stage, existing methods often adjust deviations in only one dimension, such as correcting only lateral path offset or only adjusting longitudinal speed deviation, without considering the cascading effects of correcting one dimension on another. For instance, when increasing the steering angle to correct lateral path offset, if the longitudinal speed is not adjusted simultaneously, it may lead to an imbalance in the distribution of lateral and longitudinal forces, further amplifying the trajectory deviation and reducing the overall accuracy of trajectory planning.

[0005] With the expansion of autonomous vehicle application scenarios, from structured roads to unstructured roads, environmental complexity and uncertainty have increased. Existing trajectory planning methods lack a unified lateral and longitudinal coordination mechanism when facing diverse scenarios, requiring separate adjustments to control strategies for different scenarios. This increases system development difficulty and maintenance costs, and also makes it difficult to guarantee consistent trajectory planning performance across different scenarios. These problems restrict the dynamic adaptability and safety of autonomous driving trajectory planning, necessitating a trajectory planning method that can achieve lateral and longitudinal coordination and dynamically respond to environmental changes. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination, the method comprising: Collect lateral and longitudinal motion state information of autonomous vehicles, and obtain dynamic information about the surrounding environment; Based on the lateral and longitudinal motion state information, the lateral and longitudinal coordination index is calculated. The surrounding environment dynamic information is input into a preset dynamic trajectory model to obtain reference lateral trajectory parameters and reference longitudinal trajectory parameters. The trajectory planning deviation value is determined based on the deviation between the horizontal and vertical coordination index and the reference horizontal trajectory parameters and the reference vertical trajectory parameters; Based on the trajectory planning deviation value, the lateral control parameters and longitudinal control parameters in the trajectory generation process are corrected.

[0008] Preferably, establishing the preset dynamic trajectory model includes the following steps: Collect multi-source historical driving data, including vehicle lateral displacement sequences and longitudinal speed sequences, and perform time synchronization and noise filtering on the multi-source historical driving data; The processed historical driving data is categorized by scenario type, including urban roads, highways, and curves. For each scene type, the temporal correlation of horizontal motion features and vertical motion features is extracted, and the mutual information value between features is calculated to quantify the horizontal and vertical coupling strength. Based on the mutual information value, a nonlinear mapping function between the lateral trajectory parameters and the longitudinal trajectory parameters is constructed. The parameters of the nonlinear mapping function are adjusted by an iterative optimization algorithm to minimize the error between the predicted trajectory and the historical trajectory.

[0009] Preferably, correcting the lateral and longitudinal control parameters based on the trajectory planning deviation value includes the following steps: The difference between the trajectory planning deviation value and the dynamic deviation threshold is calculated in real time, wherein the dynamic deviation threshold is adaptively adjusted according to the current driving environment; When the difference is near zero, the lateral control parameters and the longitudinal control parameters remain unchanged. When the difference deviates from zero and exceeds the predetermined tolerance, the parameter adjustment process is initiated: first, the deviation change rate is calculated, and then the preset parameter adjustment table is searched according to the deviation change rate to obtain the parameter adjustment weight; The parameter adjustment weights are applied to the incremental update formulas of the lateral and longitudinal control parameters to generate the corrected parameters.

[0010] Preferably, the process of finding the preset parameter adjustment table based on the deviation change rate includes the following steps: The preset parameter adjustment table stores the adjustment weight coefficients corresponding to multiple sets of deviation change rate intervals; Monitor the current deviation change rate and determine the deviation change rate interval to which the current deviation change rate belongs based on the parameter adjustment table; wherein the adjustment weight coefficient is positively correlated with the deviation change rate.

[0011] Preferably, determining the trajectory update interval includes the following steps: The current vehicle status is obtained, including lateral velocity and longitudinal acceleration, as well as environmental complexity indicators; The vehicle dynamic response time constant is calculated based on lateral velocity and longitudinal acceleration. Based on the environmental complexity index and dynamic response time constant, the baseline update interval is queried using interpolation. A safety margin factor is introduced to scale the baseline update interval, resulting in the final trajectory update interval.

[0012] Preferably, adjusting the trajectory update interval according to environmental changes includes the following steps: Continuously monitor environmental sensor data streams, including distance and relative speed to obstacles ahead; Calculate the gradient of environmental sensor data changes within adjacent monitoring periods; If the gradient change is below the sensitivity threshold, the trajectory update interval remains unchanged. If the gradient change exceeds the sensitivity threshold, interval adjustment is initiated: the interval scaling factor is calculated based on the magnitude and direction of the gradient change. The adjusted trajectory update interval is obtained by multiplying the trajectory update interval by the interval scaling factor.

[0013] Preferably, calculating the interval scaling factor includes the following steps: Multiple gradient ranges are preset, and each range corresponds to a basic scaling factor; Identify the range of the current gradient change and select the corresponding base scaling factor; By combining historical interval adjustment records, a moving average algorithm is applied to smooth the basic scaling factor; The interval scaling factor is finally determined based on the smoothed base scaling factor and the current vehicle speed.

[0014] Preferably, determining the trajectory parameters corresponding to each trajectory update interval includes the following steps: Record the historical trajectory update interval sequence and its corresponding lateral and longitudinal control parameters; Compare the current trajectory update interval with the nearest interval value in the historical sequence; If the current interval matches a historical interval, the historical parameter value is used directly. If the current interval is greater than the historical interval, then parameter prediction is initiated: based on the trend of historical parameter values, the current parameter estimate is calculated using linear extrapolation. The parameter estimates are normalized to ensure that the parameters are within the physically feasible range.

[0015] Preferably, calculating the current parameter estimate using linear extrapolation includes the following steps: The parameter values ​​corresponding to the three most recent historical trajectory update intervals are selected as sample points; Calculate the first difference of the parameter value over time to obtain the slope of the parameter change; Using the current time point as a reference, extend along the slope of parameter change to obtain the predicted parameter values; The Kalman filter is applied to denoise and optimize the parameter predictions to generate the final parameter estimates.

[0016] Preferably, generating the final dynamic trajectory includes the following steps: Initialize the trajectory point sequence, starting from the current vehicle position; For each trajectory update interval, load the corresponding lateral control parameters and longitudinal control parameters; The lateral displacement increment is calculated based on the lateral control parameters, and the longitudinal displacement increment is calculated based on the longitudinal control parameters. By integrating the lateral and longitudinal displacement increments, candidate trajectory points are generated; the candidate trajectory points are then smoothly connected using a spline interpolation algorithm to output a continuous dynamic trajectory.

[0017] Compared with the prior art, the beneficial effects of the present invention are: By constructing a lateral and longitudinal coordination analysis mechanism, this method breaks through the limitations of traditional separate trajectory planning, enabling the vehicle's lateral and longitudinal movements to form an organic synergy. During trajectory planning, this method does not treat lateral and longitudinal movements as independent components, but rather captures the correlation between them in real time by calculating a lateral and longitudinal coordination index. For example, in lane-changing scenarios, the lateral and longitudinal coordination index reflects the degree of matching between the current steering angle and vehicle speed. If an increase in steering angle leads to an increase in lateral acceleration, the coordination index will synchronously reflect this change, providing a basis for subsequent trajectory parameter adjustments. This avoids vehicle attitude instability caused by a disconnect between lateral and longitudinal movements, allowing the trajectory generation to better match the physical characteristics of vehicle motion, resulting in a smoother driving process and reducing ride discomfort caused by a lack of coordination between lateral and longitudinal movements.

[0018] This method enhances the responsiveness of trajectory planning to environmental changes by deeply utilizing dynamic information from the surrounding environment through a pre-set dynamic trajectory model. Before trajectory generation, the method comprehensively collects dynamic information about the surrounding environment, including obstacle movement speed, vehicle driving status in adjacent lanes, and traffic flow trends, and inputs this dynamic information into the pre-set model in real time. The pre-set dynamic trajectory model can dynamically update the reference lateral and longitudinal trajectory parameters based on real-time environmental data, rather than relying on static constraints. For example, in congested road sections, if the density of vehicles ahead suddenly increases, the model will adjust the reference longitudinal speed parameter according to real-time traffic flow data, reducing the driving speed while simultaneously optimizing the lateral trajectory parameters. This avoids frequent acceleration, deceleration, or steering adjustments caused by a mismatch between vehicle speed and the environment, ensuring that the reference trajectory always remains consistent with the actual environment, reducing trajectory deviations caused by environmental changes, and improving the dynamic adaptability of trajectory planning.

[0019] In the trajectory deviation correction stage, this method achieves precise correction of control parameters by comprehensively analyzing the deviation between the lateral and longitudinal coordination indices and the reference trajectory parameters, avoiding the chain reaction problems caused by single-dimensional correction. When determining the trajectory planning deviation value, the method does not consider lateral or longitudinal deviations in isolation, but combines them with the lateral and longitudinal coordination indices to determine whether the root cause of the deviation stems from an imbalance in lateral and longitudinal motion. For example, when a lateral trajectory deviation is detected, it first analyzes whether the excessive longitudinal vehicle speed leads to untimely steering adjustments. If such a correlation exists, the correction process simultaneously adjusts the lateral steering parameters and longitudinal speed parameters, rather than only correcting the lateral deviation. This comprehensive correction method avoids new deviations caused by single-dimensional correction, ensures that the adjustments of lateral and longitudinal control parameters are mutually compatible, improves the effectiveness of trajectory correction, enables the vehicle to quickly return to a reasonable trajectory, and reduces the impact of accumulated trajectory deviation on driving safety.

[0020] This method eliminates the need for separate control strategies for different application scenarios. Through a unified lateral and longitudinal coordination mechanism and dynamic trajectory model, it can adapt to various driving scenarios. In structured roads (such as highways), the method can coordinate lateral and longitudinal parameters based on stable traffic flow information to maintain stable high-speed driving. In unstructured roads (such as urban residential roads), it can quickly adjust lateral and longitudinal control parameters by capturing dynamic information of pedestrians and non-motorized vehicles in real time, ensuring driving safety. This scenario adaptability reduces system development and maintenance costs while ensuring consistency in trajectory planning performance across different scenarios, thus improving the method's practicality and scalability.

[0021] This method reduces conflicts during vehicle operation by calculating lateral and longitudinal coordination indices in real time and dynamically adjusting control parameters. For example, when avoiding obstacles, lateral steering and longitudinal deceleration can be synchronized, avoiding oversteering with insufficient deceleration or excessive deceleration with lag in steering, making vehicle operation more consistent and reducing safety risks caused by operational conflicts. Furthermore, consistent trajectory planning and control parameter adjustments can also reduce vehicle energy consumption. For instance, when driving on curves, lateral and longitudinal coordination adapts vehicle speed and steering angle, avoiding frequent acceleration, deceleration, or sharp turns, thus improving energy efficiency. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the working principle of the dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination described in this invention. Figure 2 A flowchart for establishing a preset dynamic trajectory model; Figure 3 A flowchart for determining the trajectory update interval. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1This invention provides a dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination. The method includes: the autonomous vehicle collecting lateral and longitudinal motion state information through an onboard sensor system. The lateral motion state information includes lateral displacement and lateral velocity, while the longitudinal motion state information includes longitudinal velocity and longitudinal acceleration. Simultaneously, an environmental perception module acquires dynamic information about the surrounding environment, such as obstacle locations, road curvature, and traffic flow data. A lateral and longitudinal coordination index is derived mathematically, involving the fusion processing of lateral and longitudinal motion state information, such as using weighted averages or covariance analysis to quantify the degree of coordination between lateral and longitudinal motion. A preset dynamic trajectory model receives the surrounding environmental dynamic information as input and outputs reference lateral trajectory parameters and reference longitudinal trajectory parameters. The reference lateral trajectory parameters include a sequence of lateral path points, and the reference longitudinal trajectory parameters include a velocity profile. The trajectory planning deviation is determined by comparing the difference between the lateral and longitudinal coordination index and the reference lateral and longitudinal trajectory parameters, such as by calculating Euclidean distance or root mean square error. The trajectory generation process involves real-time adjustment of lateral and longitudinal control parameters. Lateral control parameters include steering angle gain, and longitudinal control parameters include throttle opening. Based on the trajectory planning deviation value, a feedback control mechanism is used for correction, such as a proportional-integral-derivative controller or model predictive control algorithm, to ensure trajectory smoothness and safety.

[0025] Example 1: See Figure 2 The process of collecting multi-source historical driving data relies on a distributed onboard data recording system integrated into the electronic control unit network of the autonomous vehicle. This multi-source historical driving data includes vehicle lateral displacement sequences and longitudinal velocity sequences. The lateral displacement sequences are output by a combination of a high-precision GPS receiver and an inertial measurement unit, while the longitudinal velocity sequences are acquired through wheel speed sensors and bus signals from the engine control unit. Data acquisition takes place on a real road test track covering various typical road conditions, including urban centers, intercity highways, and mountain curves. The data collection vehicles are equipped with a complete sensor suite, and their sampling frequency is uniformly set to 100 Hz to ensure the temporal integrity of the data sequences. During the recording process, multi-source historical driving data is accompanied by timestamps accurate to the millisecond level, generated synchronously by a precision clock source within the vehicle network.

[0026] Time synchronization and noise filtering of multi-source historical driving data are performed in a dedicated data preprocessing module. Time synchronization utilizes a timestamp alignment algorithm based on a network time protocol to adjust asynchronous data streams from different buses onto a unified time axis. Noise filtering employs a cascaded filtering strategy: the raw data first passes through a low-pass finite-length unit impulse response digital filter to suppress high-frequency noise, followed by an adaptive Kalman filter to further smooth abnormal fluctuations caused by road bumps or instantaneous sensor errors. The processed data sequence remains coherent in the time domain, and its signal quality meets the signal-to-noise ratio requirements for subsequent feature extraction. The processed historical driving data is then fed into a scene classification engine. This engine automatically labels the data based on predefined road topology and traffic flow pattern features. The classification criteria clearly categorize the data into three main types: urban road scenes, highway scenes, and curve scenes. Urban road scenes are identified by an average speed below 50 km / h and frequent traffic light start-stop events. Highway scenes are identified by continuous high-speed cruising without intersection interference. Curve scenes are identified by road segments with a curvature radius of less than 500 meters. For each classified scene type, the feature extraction module analyzes the temporal correlation of lateral and longitudinal motion features in parallel. Lateral motion features mainly include the temporal variation patterns of lateral displacement, lateral velocity, and lateral acceleration, while longitudinal motion features focus on the evolution of longitudinal velocity, longitudinal acceleration, and jerk. The calculation of temporal correlation uses a dynamic time warping algorithm to measure the similarity between time series of different lengths, thereby capturing the dynamic coupling relationship between lateral and longitudinal motion in the time dimension. Based on the correlation analysis, the mutual information value between lateral motion features and longitudinal motion features is further calculated. The calculation of the mutual information value is based on the joint probability distribution and marginal probability distribution of the feature values. The probability density function is estimated by histogram statistics. As a dimensionless scalar, the mutual information value directly quantifies the statistical dependence between lateral motion and longitudinal motion in a specific driving scenario, i.e., the strength of lateral and longitudinal coupling.

[0027] Based on the calculated mutual information values, the model building phase focuses on establishing a nonlinear mapping function between lateral and longitudinal trajectory parameters. The nonlinear mapping function is structured as a feedforward neural network with two hidden layers. Input layer nodes correspond to key lateral motion features, while output layer nodes correspond to the longitudinal trajectory parameters to be predicted. The network's training data comes from a classified scene database. A dedicated nonlinear mapping function is trained independently for each scene type. The mutual information values ​​are used as a reference for initializing connection weights during training to enhance the model's ability to fit lateral and longitudinal coupling relationships. The output of the nonlinear mapping function includes a set of parameterized trajectory descriptors, such as the curvature polynomial coefficients of the lateral path and the spline control points of the longitudinal velocity profile. Adjusting the parameters of the nonlinear mapping function through an iterative optimization algorithm is an offline learning process. The iterative optimization algorithm uses stochastic gradient descent with a momentum term, and the optimization objective is defined as minimizing the sum of Euclidean distances between the trajectory points predicted by the nonlinear mapping function and the actual trajectory points recorded in multi-source historical driving data. The training process employs early stopping to prevent overfitting; training terminates when the error on the validation set no longer decreases over multiple consecutive iterations. The parameter matrix of the resulting nonlinear mapping function is then permanently stored in the autonomous vehicle's onboard solid-state drive for use in real-time trajectory planning tasks. The entire process of building the pre-defined dynamic trajectory model emphasizes the differences in scenarios and the inherent correlation between lateral and longitudinal movements, making the generated reference trajectory more consistent with real-world driving behavior patterns.

[0028] The collection specifications for multi-source historical driving data require diversity and representativeness of data sources. The data collection vehicle fleet includes different models of passenger cars, and the driving time covers various environmental conditions such as day and night cycles and sunny and rainy weather to ensure the broad applicability of the data. The recording of vehicle lateral displacement sequences includes not only absolute coordinates but also offsets relative to lane lines, while the longitudinal speed sequences are associated with transmission gear positions and brake pedal travel signals. Time synchronization processing needs to address the issue of inconsistent data delays from different sensors. The method adopted is to add hardware interrupt timestamps to the header of data packets and reorder the data through a priority queue. The filter parameters for noise filtering are initialized according to the noise characteristic manuals provided by the sensor manufacturers and fine-tuned on actual data to achieve the best filtering effect. The accuracy of scene type classification depends on the quality of feature engineering. In addition to basic vehicle speed and road curvature, the classification engine also incorporates contextual features such as traffic density and traffic light cycles. The classification algorithm itself is a support vector machine model, which is trained using a large amount of labeled road scene data. The temporal correlation analysis of lateral and longitudinal motion features employs a sliding window mechanism. The window length is dynamically adjusted according to the scene: a shorter window is used in urban road scenarios to capture frequent changes, while a longer window is used in highway scenarios to smooth long-term trends. The probability density estimation in the mutual information value calculation process uses the Gaussian kernel density estimation method, which provides a smoother probability density estimate compared to the histogram method, thus yielding a more stable mutual information value.

[0029] The feedforward neural network structure with a nonlinear mapping function is carefully designed. The ReLU activation function is used in the hidden layer to alleviate the gradient vanishing problem, while the linear activation function is used in the output layer to adapt to the parameter regression task. 70% of the training data is used for training, 15% for validation, and the remaining 15% for testing. The batch size during training is set to 128, and the learning rate adopts an exponential decay strategy. The momentum coefficient in the iterative optimization algorithm is set to 0.9 to accelerate convergence and reduce oscillations. The entire model building process is encapsulated into an automated toolchain, requiring no manual intervention from raw data input to model deployment, ensuring the consistency and repeatability of the preset dynamic trajectory model. The final generated preset dynamic trajectory model can quickly query and output reference trajectory parameters that conform to the characteristics of the current scene and are coordinated in both horizontal and vertical motion based on real-time perceived dynamic information of the surrounding environment. The quantification results of the horizontal and vertical coupling strength are stored in a multi-dimensional lookup table. The index of the lookup table is the scene type and the discretized interval of the mutual information value, and the table entries are the corresponding typical horizontal and vertical motion pattern descriptors. The parameter tuning process of the nonlinear mapping function is logged in detail, including the loss function value and parameter gradient norm for each iteration. These logs are used to monitor the health of the training process. Before deployment, the pre-defined dynamic trajectory model undergoes a rigorous set of validation test cases, simulating extreme driving scenarios to evaluate the model's robustness. The computational resources and memory budget consumed by the model during runtime are strictly limited to ensure its real-time operation on the automotive embedded platform.

[0030] Example 2: The difference between the real-time calculated trajectory planning deviation value and the dynamic deviation threshold is the trigger condition for the parameter correction process. The trajectory planning deviation value comes from the real-time output of the trajectory planning module, while the dynamic deviation threshold is a variable that changes with the environment. The adaptive adjustment mechanism of the dynamic deviation threshold relies on the multimodal data fusion of the environmental perception system. The environmental perception system integrates data from millimeter-wave radar, lidar, and vision sensors. The calculation model of the dynamic deviation threshold adopts an expert system based on fuzzy inference. This system takes the distance to the obstacle ahead, relative speed, road adhesion coefficient, and visibility as input variables, and outputs a real-time dynamic deviation threshold suggestion value through a set of fuzzy rule bases. The calculation of the difference is performed in each control cycle, and the frequency of the control cycle is synchronized with the vehicle bus cycle. The calculation result is a signed numerical value, and its positive or negative sign reflects the direction of deviation of the trajectory planning deviation value relative to the dynamic deviation threshold. When the absolute value of the difference is within a dead zone near a preset zero value, the lateral control parameters and longitudinal control parameters remain unchanged. The dead zone range is set to suppress frequent parameter oscillations caused by sensor noise or small path fluctuations. Lateral control parameters include the proportional gain and derivative gain of the steering system, while longitudinal control parameters include the torque mapping coefficient of the drive motor and the response curve parameters of regenerative braking. During the parameter holding phase, the control system enters a steady-state maintenance mode, where the controller's integral term is frozen to prevent integral saturation, and all system resources are focused on monitoring environmental changes rather than actively adjusting. Once the absolute value of the difference exceeds a predetermined tolerance, the parameter adjustment process is immediately activated. The predetermined tolerance value is dynamically set according to the vehicle's current driving mode; for example, a more lenient tolerance is used in highway cruise mode, while a very strict tolerance is used in automatic parking mode. The first step in the parameter adjustment process is to calculate the trend of change in the trajectory planning deviation value, i.e., the deviation change rate. The deviation change rate is obtained by processing the most recent trajectory planning deviation value sequence through a first-order differential filter, with the filter window length matched to the vehicle's dynamic response characteristics. The calculated deviation change rate is a scalar with physical dimensions; its magnitude and sign together describe the rate of deterioration or improvement of the trajectory tracking error.

[0031] Based on the real-time calculated deviation change rate, the system queries a preset parameter adjustment table to obtain the corresponding parameter adjustment weights. This preset parameter adjustment table is a two-dimensional lookup table stored in non-volatile memory. One dimension of the table represents the discretized intervals of the deviation change rate, divided on a logarithmic scale to cover a wide dynamic range. The other dimension corresponds to different control parameter types. Each table cell stores a pair of adjustment weight coefficients, used for correcting lateral and longitudinal control parameters respectively. The design principle for the adjustment weight coefficients is a positive correlation with the absolute value of the deviation change rate; a larger deviation change rate leads to a larger adjustment weight coefficient, allowing the controller to respond more strongly to rapidly deteriorating trajectory tracking errors. The application of parameter adjustment weights involves a set of incremental update formulas. The incremental update formula for lateral control parameters is based on the state-space model of the steering system, while the incremental update formula for longitudinal control parameters originates from the vehicle's longitudinal dynamics equations. Parameter adjustment weights are introduced as gain factors into these update formulas. The new value of the lateral control parameter is the sum of the old value and the weighted lateral deviation increment, and the new value of the longitudinal control parameter is the sum of the old value and the weighted longitudinal deviation increment. The weighted process introduces a nonlinear mapping to ensure that the parameter adjustment range is coordinated with the current vehicle state; for example, the adjustment of steering gain is more gradual when driving at high speeds. After range verification and rate limit checks, the corrected parameters are downloaded in real-time to the underlying steering controller and powertrain controller for execution. The preset parameter adjustment table is constructed based on extensive offline simulation and real-vehicle test data. The construction process employs a system identification method, injecting different types of trajectory deviation excitation signals into the vehicle model to observe the system's closed-loop response characteristics under different control parameter adjustment weights. An optimization algorithm is used to search for the adjustment weight coefficients that optimize the overall system performance, taking into account trajectory tracking accuracy, ride comfort, and actuator energy consumption. The final preset parameter adjustment table will have different versions for different vehicle platforms and chassis configurations, and is permanently embedded into the control unit using a flashing tool before the vehicle leaves the factory.

[0032] The monitoring of the deviation change rate employs a separate high-priority task, running at a frequency higher than the main control cycle to ensure timely capture of rapid changes in trajectory planning deviation values. The smoothness of the deviation change rate calculation results is guaranteed by a moving average filter, the order of which is adaptively adjusted based on the currently estimated signal-noise level. The query operation of the preset parameter adjustment table is highly optimized, employing a binary search algorithm to minimize computational latency. The query results are also interpolated to eliminate the step effect caused by discretization interval division. The integrity of the parameter adjustment process is continuously monitored. The monitoring mechanism checks the validity of the entire data link from deviation calculation to parameter download; any timeout or data anomaly in any link triggers a process reset and switches to degraded control mode. Degraded control mode uses a fixed, conservative set of control parameters to ensure basic vehicle handling stability. The entire parameter correction mechanism based on trajectory planning deviation values ​​and dynamic deviation thresholds constitutes a fast closed-loop adaptive system. This system enables autonomous vehicles to dynamically optimize lateral and longitudinal control parameters when facing complex dynamic environments, maintaining coordinated trajectory tracking performance in both directions. The adaptive adjustment of the dynamic deviation threshold is an ongoing process. The adjustment algorithm reassesses environmental factors at fixed time intervals, the length of which is consistent with the data update rate of the perception system. The evaluation process is based on a multi-factor weighted scoring model. Each environmental factor calculates a sub-score based on its current perceived value and an ideal reference value. All sub-scores are weighted and summed to obtain a comprehensive environmental risk score. This risk score is ultimately converted into a specific value for the dynamic deviation threshold through a linear mapping function. The higher the risk score, the more stringent the dynamic deviation threshold is set, thus forcing the control system to react to smaller trajectory deviations. The application of parameter adjustment weights to lateral and longitudinal control parameters is not entirely symmetrical. The adjustment of lateral control parameters focuses more on maintaining the lateral stability of the vehicle, so the application of weights takes into account the effects of tire side stiffness and load transfer. The adjustment of longitudinal control parameters prioritizes ensuring longitudinal safety distance, and the application of weights is coordinated with the strategy of the adaptive cruise control system. Before the corrected parameters take effect, they undergo a round of simulation verification on a simplified vehicle model. If the simulation results predict that parameter changes will cause the vehicle state to exceed the stability boundary, the correction magnitude will be reduced proportionally. This cautious, multi-layered parameter correction strategy ensures that the dynamic trajectory planning system for autonomous vehicles always prioritizes vehicle dynamics stability while adapting to environmental changes.

[0033] Example 3: See Figure 3The process of determining the trajectory update interval begins with continuous perception of the current state of the autonomous vehicle. The current vehicle state includes the lateral velocity obtained from lateral displacement differential and the longitudinal acceleration directly measured by the longitudinal accelerometer. These data are updated in real time via the vehicle bus at a frequency of 100 Hz. The acquisition of the environmental complexity index relies on a multi-sensor fusion module, which integrates the outputs of LiDAR, millimeter-wave radar, and vision sensors. The environmental complexity index is calculated by weighted fusion of the number, type, motion uncertainty of dynamic obstacles within a 100-meter radius ahead, and the rate of change of lane curvature, ultimately producing a normalized scalar between zero and one; a higher value indicates a more complex environment. Estimating the vehicle's dynamic response time constant requires establishing a model reflecting the vehicle's lateral and longitudinal coupling dynamics. This model treats the vehicle as a mass block, whose response delay is closely related to its current motion state. The mathematical expression for estimating the vehicle's dynamic response time constant based on lateral velocity and longitudinal acceleration is as follows:

[0034] in: This represents the estimated vehicle dynamic response time constant, with the dimension of seconds. It is a dimensionless vehicle-specific calibration coefficient. This represents the real-time lateral velocity, measured in meters per second. This represents the longitudinal velocity, which is acquired synchronously with the lateral velocity, and is measured in meters per second. This represents the longitudinal acceleration acquired in real time, measured in meters per second squared. This represents the estimated coefficient of adhesion between the tire and the road surface obtained through the road surface estimation system. This represents the gravitational acceleration constant.

[0035] Vehicle dynamic response time constant The numerator combines lateral and longitudinal velocities, using the magnitude of the velocity vector to characterize the vehicle's current overall kinetic energy state. The denominator combines longitudinal acceleration and the maximum available acceleration estimated based on the road adhesion coefficient, characterizing the vehicle's currently available longitudinal dynamic potential. The ratio describes the timescale required for the vehicle to change from its current state of motion, and the calibration coefficient... Inertial characteristics adapted to different vehicle models. Based on environmental complexity indicators and vehicle dynamic response time constant. The system queries a pre-generated two-dimensional lookup table to obtain the baseline update interval. This lookup table has been calibrated using simulation and test data covering various typical and extreme scenarios before the vehicle control unit leaves the factory. The query process uses a bilinear interpolation algorithm to ensure a smooth transition of the baseline update interval value within the continuous variation range of environmental complexity indicators and vehicle dynamic response time constants. The introduced safety margin factor is a scaling factor based on real-time risk assessment. The risk assessment module comprehensively considers weather conditions, sensor confidence, and system redundancy. The safety margin factor is always greater than or equal to one. The baseline update interval is multiplied by the safety margin factor to obtain the final trajectory update interval.

[0036] Continuous monitoring of environmental sensor data streams is a high-priority background task. This data stream includes obstacle distance data from the forward-facing main radar and lane curvature data from the vision system, updated at a frequency of 50 Hz. The system calculates the gradient of environmental sensor data changes within adjacent monitoring periods. The length of the monitoring period is independent of the trajectory update interval. Gradient calculation is performed for each key environmental parameter; for example, the gradient of obstacle distance changes is the difference between the current value and the previous period's value divided by the monitoring period length. The system presets a configurable sensitivity threshold, which is positively correlated with the vehicle's current speed. A smaller threshold is used at high speeds to improve the system's sensitivity to environmental changes. If the absolute value of the calculated gradient of an environmental parameter change is below the sensitivity threshold, the system's decision logic determines that the environment is relatively stable, the current trajectory update interval remains unchanged, and the trajectory planning module continues to operate at its original pace. If the absolute value of the gradient of any key environmental parameter change exceeds its corresponding sensitivity threshold, a dynamic adjustment process for the trajectory update interval is immediately triggered. The adjustment process calculates an interval scaling factor based on the magnitude and direction of the gradient change exceeding the limit. The direction of the gradient indicates whether the environment is becoming more complex or simpler. The interval scaling factor is calculated based on a predefined monotonic function, ensuring that the factor's magnitude is positively correlated with the absolute value of the gradient change. The calculated interval scaling factor is multiplied by the current trajectory update interval to produce an adjusted new interval value. This new interval value is constrained between a preset minimum and maximum value to prevent excessively high update frequencies from overloading computational resources or excessively low frequencies from degrading control performance.

[0037] The quantification algorithm for environmental complexity is adaptive; it assigns different risk weights to different types of obstacles, for example, giving higher weights to vehicles cutting into the lane than to vehicles traveling in the same direction. Vehicle dynamic response time constant. Calibration coefficients in the formula Different types of vehicle chassis have different values; sports chassis typically correspond to smaller values. Value, comfort chassis corresponds to larger The generation of the two-dimensional lookup table is an offline optimization process. The optimization objective is to maximize the average update interval while ensuring trajectory tracking performance to save computational resources. The dynamic adjustment of the safety margin factor is related to the vehicle's driving mode. In aggressive driving mode, the safety margin factor is lower to pursue performance, while in conservative driving mode, the safety margin factor is higher to prioritize safety. Monitoring of environmental sensor data streams not only focuses on the area in front but also periodically scans data from side and rear sensors to detect potential blind spot hazards. The calculation of changing gradients uses a third-order Savitzky-Gore filter for smoothing to suppress gradient jitter caused by sensor measurement noise. The mapping relationship between the interval scaling factor and the changing gradient is designed as a smooth S-curve. This design allows for gentle adjustment when the gradient is slightly above the threshold and rapid but saturated adjustment when the gradient is much above the threshold, avoiding drastic jumps in the update interval. The response speed of the entire trajectory update interval adjustment mechanism has been carefully tuned to be faster than environmental changes but slower than the inherent noise frequency of the sensors, thus achieving a balance between responsiveness and stability. The adjusted trajectory update interval is synchronously sent to the trajectory planning module and the control system. The trajectory planning module adjusts its planning time domain and optimization step size accordingly, while the control system adjusts its control cycle to ensure the entire system operates in coordination under a unified time reference. This adaptive interval management strategy enables the autonomous driving system to operate efficiently in simple road environments and with stable vehicle conditions, while increasing the control frequency to ensure safety and accuracy in complex and dynamic environments. Historical interval adjustment records and corresponding environmental data are recorded for offline analysis to further optimize lookup tables and adjust parameters.

[0038] Example 4: Pre-setting multiple gradient ranges forms the basis for calculating the interval scaling factor. These gradient ranges are divided based on the statistical distribution of a large amount of historical environmental sensor data streams. The boundary values ​​of the gradient ranges are set using a logarithmic scale to cover a wide dynamic range from gradual changes to abrupt changes. Each gradient range corresponds to a base scaling factor, determined through offline optimization. The optimization goal is to ensure that the adjustment of the trajectory update interval smoothly matches the severity of environmental changes. Identifying the range to which the current gradient belongs is a pattern matching process. The system compares the real-time calculated gradient value with the boundaries of the multiple pre-set gradient ranges. The range identification algorithm uses an efficient interval tree data structure for fast lookup. Selecting the corresponding base scaling factor is a step executed immediately after successfully identifying the gradient range. The base scaling factor is read from a predefined mapping relationship. The base scaling factor is smoothed by combining historical interval adjustment records, which are stored in a fixed-length first-in-first-out queue. The queue stores the base scaling factors used in the most recent interval adjustments and their corresponding timestamps. A moving average algorithm is applied to smooth the base scaling factor. The window size of the moving average algorithm is consistent with the length of the historical interval adjustment record queue. The algorithm calculates the arithmetic mean of all base scaling factors within the window, and uses this average as the smoothed base scaling factor. The smoothing process effectively suppresses the interference caused by abnormal single gradient calculations due to instantaneous sensor noise on the interval adjustment.

[0039] The interval scaling factor is finally determined based on the smoothed base scaling factor and the current vehicle speed, with the current vehicle speed being an important adjustment variable introduced into the calculation. The higher the vehicle speed, the faster the system should respond to environmental changes, and the greater the correction magnitude of the interval scaling factor to the base scaling factor. The determination process involves a speed sensitivity coefficient, which maps the physical quantity of vehicle speed to a weight. The final interval scaling factor is the product of the smoothed base scaling factor and the speed weight factor. This product relationship ensures that the interval adjustment simultaneously considers the rate of environmental change and the vehicle's own state. The interval scaling factor is constrained within a reasonable upper and lower bound to prevent unrealistic trajectory update intervals. Recording the historical trajectory update interval sequence and its corresponding lateral and longitudinal control parameters is the data preparation stage for parameter prediction. The historical trajectory update interval sequence is stored in the form of a time-series database. Each storage entry contains the value of the trajectory update interval, the timestamp of the recording time, and the set of lateral and longitudinal control parameters actually used at that interval. The nearest neighbor search algorithm is used to compare the current trajectory update interval with the nearest interval value in the historical sequence. The algorithm searches for one or more records in the historical trajectory update interval sequence that are closest to the current trajectory update interval value. The proximity is measured using Euclidean distance, and the search scope is limited to data within a recent period to avoid using overly old records.

[0040] Referring to Table 1, when the current trajectory update interval successfully matches the nearest interval value in the historical sequence, the criterion for a successful match is that the difference between the current interval and the historical interval is less than a preset tolerance threshold. The system directly uses the lateral and longitudinal control parameters stored in the historical records as the parameter values ​​for the current interval. This direct reuse strategy is based on the assumption of environmental similarity, which holds that similar trajectory update intervals often correspond to similar control parameter requirements. When the current trajectory update interval is greater than any interval value in the historical sequence, the system initiates the parameter prediction process. Parameter prediction is based on the changing trend of historical parameter values, using linear extrapolation to calculate the current parameter estimate. Linear extrapolation assumes that the control parameters change linearly over time, an assumption that is generally reasonable on short time scales. Normalizing the parameter estimate is the final step in parameter prediction, ensuring that the parameter estimate falls within the physically feasible range. Each control parameter has its physical limits; for example, there is an upper limit to the steering angle gain to prevent oversteering, and an upper limit to the acceleration command to ensure ride comfort. Normalization is achieved through a pruning function; any parameter estimate exceeding the preset minimum and maximum range will be automatically adjusted to the boundary value. The processed parameter estimates are sent to the trajectory planning module for execution and are also added to the historical records for future prediction reference.

[0041] Table 1: Mapping Table of Variation Gradient Range and Basic Scaling Factor

[0042] The gradient range and basic scaling factor mapping table defines the basic mapping relationship from the environmental change gradient to the interval adjustment intensity. The lower and upper limits of the gradient range represent the severity level of the environmental change. A basic scaling factor less than one indicates that the trajectory update interval needs to be shortened to respond to environmental changes more quickly, while a basic scaling factor greater than one indicates that the trajectory update interval can be appropriately extended to save computational resources. The mapping table is constructed based on extensive simulation tests and real vehicle data to ensure that the scaling factor within each gradient interval maintains good control performance under the corresponding environment. The table data is stored in the non-volatile memory of the vehicle control unit and is loaded into memory for fast retrieval during system runtime. The length of the historical interval adjustment record queue is a configurable parameter, and the length setting needs to strike a balance between memory length and adaptability. A queue that is too short cannot effectively smooth noise, while a queue that is too long will cause the system to respond too slowly to environmental changes. The weights of the moving average algorithm can be designed as uniform weights or exponential weights. Uniform weights treat all historical records equally, while exponential weights give higher weights to recent records. The adjustment of the interval scaling factor by vehicle speed is implemented through a speed weight factor function, which is usually a monotonically increasing function of speed. The higher the current vehicle speed, the more the speed weighting factor deviates from one, resulting in a greater correction magnitude of the interval scaling factor to the base scaling factor. The linear extrapolation method for parameter prediction requires selecting appropriate historical data points, typically the parameter values ​​corresponding to the three most recent trajectory update intervals. The first-order difference of the parameter value over time is calculated to obtain the parameter change slope. Using the current time point as a reference, the parameter change slope is extended to obtain the predicted parameter value. A Kalman filter is applied to denoise and optimize the predicted parameter value. The Kalman filter utilizes the dynamic model of parameter changes and the statistical characteristics of measurement noise to generate smoother and more reliable final parameter estimates. The boundary values ​​for normalization constraint processing are derived from vehicle dynamic limits. These boundary values ​​can be set differently for different vehicle models and driving modes to adapt to different performance requirements. The entire parameter prediction and adjustment process constitutes an adaptive control parameter management system. This system can automatically adjust the lateral and longitudinal control parameters according to changes in the trajectory update interval, maintaining the control performance of the autonomous driving system at different update frequencies. Historical record maintenance includes periodically cleaning up expired data to prevent the database from expanding indefinitely and affecting query efficiency. The results of each parameter adjustment are logged for subsequent performance analysis and system optimization.

[0043] Example 5: The initialization of the trajectory point sequence is the starting point for dynamic trajectory generation. A fixed-size buffer is allocated in memory for the trajectory point sequence, with the current vehicle position as the starting point. The current vehicle position is obtained by fusing GPS signals, inertial measurement unit data, and wheel speed-oddscopy information. The positioning data includes three-dimensional coordinates and heading angle. The initialization operation of the trajectory point sequence clears the historical data in the buffer and writes the current vehicle position as the first trajectory point. The timestamp of the first trajectory point is marked with the system time at the time of generation. The data structure design of the trajectory point sequence supports efficient insertion and deletion operations to adapt to dynamic trajectory updates. For each trajectory update interval, the system loads the corresponding lateral and longitudinal control parameters from the parameter management module. The lateral control parameters include the target curvature gain of the front wheel steering angle, and the longitudinal control parameters include the target acceleration command. The parameter loading process involves one memory read operation, with the read address indexed by the current trajectory update interval value. The loaded lateral and longitudinal control parameters are temporarily stored in the processor's registers for subsequent calculations, and the validity of the parameters is verified and range checked.

[0044] Calculating the lateral displacement increment based on lateral control parameters involves a vehicle kinematics model, which is simplified using a bicycle model. The model treats the vehicle as a rigid body, with the front axle center as the reference point. The lateral displacement increment is determined by the target curvature, wheelbase, and current longitudinal velocity in the lateral control parameters. The calculation process solves a geometric relationship to obtain the displacement change of the vehicle in the direction perpendicular to the current heading within the trajectory update interval. Calculating the longitudinal displacement increment based on longitudinal control parameters uses a uniform acceleration kinematic formula, where the longitudinal displacement increment is determined by the target acceleration and current longitudinal velocity in the longitudinal control parameters. The formula integrates the effect of the target acceleration within the trajectory update interval to obtain the displacement change in the longitudinal direction. The displacement increment calculation considers the actuator response delay, correcting the ideal command by introducing a first-order hysteresis loop. Fusing the lateral and longitudinal displacement increments to generate candidate trajectory points is a coordinate transformation process, with the previous trajectory point as the origin of the reference coordinate system. The lateral displacement increment is mapped to the lateral axis of the reference coordinate system, and the longitudinal displacement increment is mapped to the longitudinal axis. The coordinate transformation uses a two-dimensional rotation matrix, with the rotation angle being the heading angle of the previous trajectory point. The transformed lateral and longitudinal displacements are vector-added, and the result is superimposed on the coordinates of the previous trajectory point to generate a new 3D spatial point. The heading angle of the new point is updated by considering the rate of change of heading caused by lateral motion, generating complete candidate trajectory points. Each candidate trajectory point contains position coordinates, heading angle, timestamp, and corresponding velocity information.

[0045] Candidate trajectory points are smoothly connected using a spline interpolation algorithm, which selects cubic B-spline curves. The algorithm uses a series of discrete candidate trajectory points as control points to fit a smooth path with continuous second-order derivatives. Generating the spline curve requires solving a tridiagonal linear equation system to determine the B-spline basis function coefficients for each control point. The solution process uses a chasing method to ensure numerical stability; the fitted curve passes through the candidate trajectory points but is not forced to cross them, allowing the curve to smoothly approximate the point sequence. The spline interpolation algorithm also applies curvature constraints to ensure that the curvature of the generated path is continuous and does not exceed the vehicle's mechanical limits. The smoothed path point sequence is uniformly resampled in the time dimension, with the resampling frequency consistent with the control system update time. Outputting the continuous dynamic trajectory involves transmitting the path point sequence from memory to the underlying controller. The data format of the continuous dynamic trajectory conforms to the autonomous driving system interface specification. The transmission protocol uses a high-priority real-time data bus to ensure low-latency delivery of trajectory data. The trajectory point sequence includes timestamps and confidence information, which the underlying controller uses for look-ahead control. After the output is completed, the trajectory generation module enters an idle state, waiting for the next trajectory update interval to trigger a new calculation cycle.

[0046] To illustrate the above process using a specific urban road following scenario, an autonomous vehicle travels at 40 kilometers per hour behind the vehicle in front. The current vehicle position is determined by the integrated navigation system, with coordinate accuracy down to the centimeter level. The trajectory update interval is set to 100 milliseconds based on environmental complexity. The parameter management module loads a set of optimized control parameters at 100-millisecond intervals. The lateral control parameters focus on maintaining the tracking accuracy of the lane centerline, while the longitudinal control parameters focus on maintaining a safe following distance. Based on the lateral control parameters, the lateral displacement increment is calculated. The vehicle kinematics model, based on the current speed of 40 kilometers per hour and the target curvature gain, calculates that the vehicle needs to make a slight adjustment of 0.03 meters to the left within the next 100 milliseconds to conform to the road curvature. Based on the longitudinal control parameters, the longitudinal displacement increment is calculated. The longitudinal motion model, based on the target acceleration command, calculates that the vehicle needs to advance 1.12 meters along the current heading within 100 milliseconds. The fusion process synthesizes the 0.03-meter lateral displacement and the 1.12-meter longitudinal displacement, using the previous vehicle position as the origin, and through coordinate rotation and vector addition, obtains a new spatial point coordinate. This new point is added to the trajectory point sequence as a candidate trajectory point. The spline interpolation algorithm uses the five most recent candidate trajectory points as control points to generate a smooth B-spline curve segment. The curve segment transitions smoothly near the candidate trajectory points, with continuous curvature changes without abrupt changes. The algorithm outputs a high-precision path of five meters in length, with path points evenly distributed at ten-centimeter intervals. This continuous dynamic trajectory is sent to the steering controller and drive controller via the controller area network bus. The steering controller calculates the target steering angle based on the path curvature, and the drive controller adjusts the throttle and brakes based on the path velocity profile.

[0047] The trajectory point sequence buffer uses a circular queue structure; when the number of trajectory points reaches the buffer limit, the oldest point is automatically overwritten. The parameter management module maintains a parameter lookup table, with entries indexed by the trajectory update interval, storing the optimal control parameters at different frequencies. The calculation of lateral displacement increment incorporates tire lateral stiffness compensation to improve the model's prediction accuracy at high speeds. The calculation of longitudinal displacement increment considers the influence of road slope, compensating for gravity components using pitch angle sensor data. The candidate trajectory point generation process includes a collision detection check to ensure that new points do not overlap with any known obstacles. The spline interpolation algorithm's control point selection strategy is adaptive, automatically increasing control point density in areas of high path curvature and decreasing the number of control points on straight sections. The node vector of the B-spline curve is non-uniformly distributed based on the control point spacing, with denser nodes placed where the path changes drastically. The output of the continuous dynamic trajectory includes covariance information of the path points, characterizing the uncertainty of trajectory prediction. The underlying controller uses this covariance information to adjust the controller's robustness parameters, employing a more conservative control strategy when uncertainty is high. The execution time of the entire trajectory generation loop has been rigorously measured to ensure completion within the trajectory update interval even under worst-case conditions. Through this systematic approach, the dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination can generate smooth, safe, and dynamically feasible driving trajectories in real time.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination, characterized in that, Includes the following steps: Collect lateral and longitudinal motion state information of autonomous vehicles, and obtain dynamic information about the surrounding environment; Based on the lateral and longitudinal motion state information, the lateral and longitudinal coordination index is calculated. The surrounding environment dynamic information is input into a preset dynamic trajectory model to obtain reference lateral trajectory parameters and reference longitudinal trajectory parameters. The trajectory planning deviation value is determined based on the deviation between the horizontal and vertical coordination index and the reference horizontal trajectory parameters and the reference vertical trajectory parameters; Based on the trajectory planning deviation value, the lateral control parameters and longitudinal control parameters in the trajectory generation process are corrected.

2. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 1, characterized in that, Establishing the preset dynamic trajectory model includes the following steps: Collect multi-source historical driving data, including vehicle lateral displacement sequences and longitudinal speed sequences, and perform time synchronization and noise filtering on the multi-source historical driving data; The processed historical driving data is categorized by scenario type, including urban roads, highways, and curves. For each scene type, the temporal correlation of horizontal motion features and vertical motion features is extracted, and the mutual information value between features is calculated to quantify the horizontal and vertical coupling strength. Based on the mutual information value, a nonlinear mapping function between the lateral trajectory parameters and the longitudinal trajectory parameters is constructed. The parameters of the nonlinear mapping function are adjusted by an iterative optimization algorithm to minimize the error between the predicted trajectory and the historical trajectory.

3. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 1, characterized in that, Correcting the lateral and longitudinal control parameters based on the trajectory planning deviation value includes the following steps: The difference between the trajectory planning deviation value and the dynamic deviation threshold is calculated in real time, wherein the dynamic deviation threshold is adaptively adjusted according to the current driving environment; When the difference is near zero, the lateral control parameters and the longitudinal control parameters remain unchanged. When the difference deviates from zero and exceeds the predetermined tolerance, the parameter adjustment process is initiated: first, the deviation change rate is calculated, and then the preset parameter adjustment table is searched according to the deviation change rate to obtain the parameter adjustment weight; The parameter adjustment weights are applied to the incremental update formulas of the lateral and longitudinal control parameters to generate the corrected parameters.

4. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 3, characterized in that, Finding the preset parameter adjustment table based on the deviation change rate involves the following steps: The preset parameter adjustment table stores the adjustment weight coefficients corresponding to multiple sets of deviation change rate intervals; Monitor the current deviation change rate and determine the deviation change rate interval to which the current deviation change rate belongs based on the parameter adjustment table; wherein the adjustment weight coefficient is positively correlated with the deviation change rate.

5. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 1, characterized in that, Determining the trajectory update interval includes the following steps: The current vehicle status is obtained, including lateral velocity and longitudinal acceleration, as well as environmental complexity indicators; The vehicle dynamic response time constant is calculated based on lateral velocity and longitudinal acceleration. Based on the environmental complexity index and dynamic response time constant, the baseline update interval is queried using interpolation. A safety margin factor is introduced to scale the baseline update interval, resulting in the final trajectory update interval.

6. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 5, characterized in that, Adjusting the trajectory update interval based on environmental changes includes the following steps: Continuously monitor environmental sensor data streams, including distance and relative speed to obstacles ahead; Calculate the gradient of environmental sensor data changes within adjacent monitoring periods; If the gradient change is below the sensitivity threshold, the trajectory update interval remains unchanged. If the gradient change exceeds the sensitivity threshold, interval adjustment is initiated: the interval scaling factor is calculated based on the magnitude and direction of the gradient change. The adjusted trajectory update interval is obtained by multiplying the trajectory update interval by the interval scaling factor.

7. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 6, characterized in that, Calculating the interval scaling factor includes the following steps: Multiple gradient ranges are preset, and each range corresponds to a basic scaling factor; Identify the range of the current gradient change and select the corresponding base scaling factor; By combining historical interval adjustment records, a moving average algorithm is applied to smooth the basic scaling factor; The interval scaling factor is finally determined based on the smoothed base scaling factor and the current vehicle speed.

8. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 5, characterized in that, Determining the trajectory parameters corresponding to each trajectory update interval includes the following steps: Record the historical trajectory update interval sequence and its corresponding lateral and longitudinal control parameters; Compare the current trajectory update interval with the nearest interval value in the historical sequence; If the current interval matches a historical interval, the historical parameter value is used directly. If the current interval is greater than the historical interval, then parameter prediction is initiated: based on the trend of historical parameter values, the current parameter estimate is calculated using linear extrapolation. The parameter estimates are normalized to ensure that the parameters are within the physically feasible range.

9. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 8, characterized in that, The steps involved in calculating the current parameter estimate using linear extrapolation are as follows: The parameter values ​​corresponding to the three most recent historical trajectory update intervals are selected as sample points; Calculate the first difference of the parameter value over time to obtain the slope of the parameter change; Using the current time point as a reference, extend along the slope of parameter change to obtain the predicted parameter values; The Kalman filter is applied to denoise and optimize the parameter predictions to generate the final parameter estimates.

10. The dynamic trajectory planning method for autonomous vehicles based on lateral and longitudinal coordination as described in claim 1, characterized in that, Generating the final dynamic trajectory includes the following steps: Initialize the trajectory point sequence, starting from the current vehicle position; For each trajectory update interval, load the corresponding lateral control parameters and longitudinal control parameters; The lateral displacement increment is calculated based on the lateral control parameters, and the longitudinal displacement increment is calculated based on the longitudinal control parameters. By integrating the lateral and longitudinal displacement increments, candidate trajectory points are generated; the candidate trajectory points are then smoothly connected using a spline interpolation algorithm to output a continuous dynamic trajectory.

Images