An unmanned vehicle driving control method and system based on dynamic obstacle avoidance

Abstract

The application provides a kind of unmanned vehicle driving control method and system based on dynamic obstacle avoidance, it is related to unmanned vehicle technical field, the method comprises: the target multi-source heterogeneous data of unmanned vehicle is collected in real time, target driving control data and target environmental disturbance data are jointly reconstructed, target candidate obstacle avoidance driving trajectory set is generated by cooperating with preset obstacle avoidance maneuver instruction set and target motion state estimation result, extract the first candidate trajectory, construct target anti-collision envelope by cooperating with the first candidate trajectory and the first position uncertainty boundary, carry out collision threat assessment, trigger target obstacle avoidance mode based on target evaluation result, determine the optimal obstacle avoidance driving trajectory to carry out the obstacle avoidance control of target unmanned vehicle.The technical problems that the state estimation of prior art is fixed, and the trajectory uncertainty modeling relies on theoretical assumptions lack of empirical support are solved.Through combining dynamic environment and historical driving experience, the technical effect of improving the safety and reliability of unmanned vehicle obstacle avoidance decision is realized.

Description

Technical Field

[0001] This invention relates to the field of unmanned vehicle technology, specifically to an unmanned vehicle driving control method and system based on dynamic obstacle avoidance. Background Technology

[0002] With the rapid development of autonomous driving technology, the ability of vehicles to autonomously avoid obstacles has become crucial to ensuring driving safety. Existing state estimation methods for autonomous vehicles typically rely on pre-set fixed dynamic models or filtering algorithms, such as state prediction based on Kalman filtering or extended Kalman filtering. These methods assume that the vehicle motion process and environmental disturbances satisfy specific linear or nonlinear models and use fixed process noise covariance and measurement noise covariance for state updates.

[0003] However, in complex real-world environments, vehicles are affected by various factors such as road conditions, lighting changes, communication interference, and traffic dynamics, resulting in highly nonlinear and random motion. Fixed models cannot reflect environmental changes in a timely manner, leading to discrepancies between predicted and actual motion states. Furthermore, trajectory uncertainty modeling relies heavily on theoretical assumptions and lacks dynamic corrections based on historical driving experience and actual environmental changes. This causes differences between generated candidate obstacle avoidance trajectories and the vehicle's actual executable trajectories, reducing the accuracy of motion state estimation and trajectory planning. This can easily lead to misjudgments in obstacle avoidance decisions, impacting the driving safety and reliability of autonomous vehicles.

[0004] Existing technologies suffer from the problem that the state estimation of autonomous vehicles is fixed and the trajectory uncertainty modeling relies on theoretical assumptions and lacks empirical support, leading to technical problems such as easy misjudgment in obstacle avoidance decisions in complex dynamic scenarios. Summary of the Invention

[0005] The purpose of this application is to provide a driving control method and system for unmanned vehicles based on dynamic obstacle avoidance, which solves the technical problem that existing technologies have fixed state estimation of unmanned vehicles and trajectory uncertainty modeling that relies on theoretical assumptions and lacks empirical support, leading to easy misjudgment of obstacle avoidance decisions in complex dynamic scenarios.

[0006] In view of the above problems, this application provides a method and system for unmanned vehicle driving control based on dynamic obstacle avoidance.

[0007] The first aspect of this application provides a driving control method for unmanned vehicles based on dynamic obstacle avoidance. This method includes: real-time acquisition of multi-source heterogeneous data of a target unmanned vehicle; joint reconstruction of target driving control data and target environmental disturbance data from the multi-source heterogeneous data to obtain a target motion state estimation result for the target unmanned vehicle; generation of a target candidate obstacle avoidance trajectory set by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and extraction of a first candidate trajectory from the target candidate obstacle avoidance trajectory set, wherein the first candidate trajectory corresponds to a first position uncertainty boundary; construction of a target collision avoidance envelope by coordinating the first candidate trajectory and the first position uncertainty boundary, and collision threat assessment by combining target obstacle perception data to obtain a target assessment result; triggering a target obstacle avoidance mode based on the target assessment result, and determining an optimal obstacle avoidance trajectory according to the obstacle avoidance fitness strategy under the target obstacle avoidance mode; and performing obstacle avoidance control of the target unmanned vehicle based on the optimal obstacle avoidance trajectory.

[0008] Optionally, the target multi-source heterogeneous data includes at least the target driving control data, the target environmental disturbance data, and the target obstacle perception data; wherein the target driving control data includes satellite positioning data, inertial measurement data, wheel speed data, and wheel steering angle data; wherein the target environmental disturbance data includes light intensity data, communication environment data, and traffic condition data; and wherein the target obstacle perception data includes obstacle size, obstacle location, and passability category.

[0009] Optionally, the target driving control data and the target environmental disturbance data are processed using an adaptive capacitive Kalman filter method to form a measurement input; the measurement input is analyzed using a preset unmanned vehicle dynamics model to obtain the target motion state estimation result; wherein, the preset unmanned vehicle dynamics model embeds a dynamic process noise covariance matrix, and the dynamic process noise covariance matrix is ​​obtained by dynamically correcting the preset process noise covariance matrix using a dynamic forgetting factor, and the dynamic forgetting factor is negatively correlated with the dynamic environmental factor.

[0010] Optionally, the measured input is analyzed by a preset unmanned vehicle dynamics model to obtain the target motion state estimation result. This includes encoding and normalizing the light intensity data, the communication environment data, and the traffic condition data, and then using a variation weighting method to obtain the dynamic environment factor.

[0011] Optionally, a two-dimensional Gaussian distribution is introduced to determine the confidence interval of the position uncertainty; the historical driving logs of the first candidate trajectory are obtained, and candidate historical trajectories with trajectory similarity reaching a predetermined threshold are selected; historical trajectories with confidence levels in the position uncertainty confidence interval are selected from the candidate historical trajectories to form the first target historical trajectory; the first target historical trajectory is used to form the first position uncertainty boundary.

[0012] Optionally, any historical trajectory is extracted from the historical driving log; the trajectory similarity between the first candidate trajectory and the arbitrary historical trajectory is evaluated to obtain an arbitrary similarity; if the arbitrary similarity is within the predetermined threshold, the arbitrary historical trajectory is added to the candidate historical trajectory.

[0013] Optionally, in the target obstacle avoidance mode, the obstacle avoidance fitness of the first candidate trajectory is evaluated to obtain a first fitness; the target candidate obstacle avoidance driving trajectory set is optimized with the first fitness as the target to determine the optimal obstacle avoidance driving trajectory; wherein, the first fitness is the normalized result of the mean trajectory similarity between each historical trajectory in the first target historical trajectory and the first candidate trajectory.

[0014] A second aspect of this application provides an unmanned vehicle driving control system based on dynamic obstacle avoidance. The system includes: a data acquisition module for real-time acquisition of multi-source heterogeneous data of the target unmanned vehicle; a joint reconstruction module for jointly reconstructing target driving control data and target environmental disturbance data from the multi-source heterogeneous data to obtain a target motion state estimation result for the target unmanned vehicle; a candidate trajectory extraction module for generating a target candidate obstacle avoidance driving trajectory set by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and extracting a first candidate trajectory from the target candidate obstacle avoidance driving trajectory set, wherein the first candidate trajectory corresponds to a first position uncertainty boundary; a collision assessment module for constructing a target collision avoidance envelope by coordinating the first candidate trajectory and the first position uncertainty boundary, and performing a collision threat assessment by combining target obstacle perception data to obtain a target assessment result; a trajectory determination module for triggering a target obstacle avoidance mode based on the target assessment result, and determining an optimal obstacle avoidance driving trajectory according to the obstacle avoidance fitness strategy under the target obstacle avoidance mode; and an obstacle avoidance control module for performing obstacle avoidance control of the target unmanned vehicle based on the optimal obstacle avoidance driving trajectory.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0016] The method provided in this application embodiment is an unmanned vehicle driving control method based on dynamic obstacle avoidance, comprising: real-time acquisition of target multi-source heterogeneous data of the target unmanned vehicle; joint reconstruction of target driving control data and target environmental disturbance data in the target multi-source heterogeneous data to obtain target motion state estimation results of the target unmanned vehicle; generating a target candidate obstacle avoidance driving trajectory set by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation results, and extracting a first candidate trajectory from the target candidate obstacle avoidance driving trajectory set, wherein the first candidate trajectory corresponds to a first position uncertainty boundary; constructing a target collision avoidance envelope by coordinating the first candidate trajectory and the first position uncertainty boundary, and performing collision threat assessment by combining target obstacle perception data to obtain a target assessment result; triggering a target obstacle avoidance mode based on the target assessment result, and determining the optimal obstacle avoidance driving trajectory according to the obstacle avoidance fitness strategy in the target obstacle avoidance mode; and performing obstacle avoidance control of the target unmanned vehicle based on the optimal obstacle avoidance driving trajectory. This achieves the technical effect of improving the safety and reliability of unmanned vehicle obstacle avoidance decision-making in complex dynamic scenarios by combining dynamic environment and historical driving experience.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an unmanned vehicle driving control method based on dynamic obstacle avoidance provided in this application.

[0020] Figure 2 This is a schematic diagram of the structure of an unmanned vehicle driving control system based on dynamic obstacle avoidance, provided in this application.

[0021] Figure labeling: Data acquisition module 11, joint reconstruction module 12, candidate trajectory extraction module 13, collision assessment module 14, trajectory determination module 15, obstacle avoidance control module 16. Detailed Implementation

[0022] This application provides a method and system for autonomous vehicle driving control based on dynamic obstacle avoidance. It addresses the technical problems in existing technologies where the autonomous vehicle's state estimation is fixed and trajectory uncertainty modeling relies on theoretical assumptions lacking empirical support, leading to proneness to misjudgments in obstacle avoidance decisions in complex dynamic scenarios. By combining dynamic environment and historical driving experience, it achieves the technical effect of improving the safety and reliability of autonomous vehicle obstacle avoidance decisions in complex dynamic scenarios.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides a method for controlling the driving of an unmanned vehicle based on dynamic obstacle avoidance. This method includes:

[0025] Real-time acquisition of multi-source heterogeneous data of the target unmanned vehicle.

[0026] Furthermore, the target multi-source heterogeneous data includes at least the target driving control data, the target environmental disturbance data, and the target obstacle perception data; wherein, the target driving control data includes satellite positioning data, inertial measurement data, wheel speed data, and wheel steering angle data; wherein, the target environmental disturbance data includes light intensity data, communication environment data, and traffic condition data; wherein, the target obstacle perception data includes obstacle size, obstacle location, and passability category.

[0027] Specifically, the multi-source heterogeneous data of the target unmanned vehicle is collected in real time by various sensors and communication modules on the vehicle. The multi-source heterogeneous data includes at least target driving control data, target environmental disturbance data and target obstacle perception data. Among them, the target driving control data is used to describe the motion state of the unmanned vehicle itself and mainly comes from the on-board sensors and control system, including satellite positioning data, inertial measurement data, wheel speed data and wheel steering angle data.

[0028] Satellite positioning data refers to the real-time acquisition of vehicle latitude, longitude, speed, and heading information through the satellite positioning module on board the unmanned vehicle, used for preliminary positioning of the vehicle's position and trajectory. Inertial measurement data includes three-axis acceleration and angular velocity data, used to supplement the short-term accuracy of satellite positioning. Wheel speed data refers to the rotational speed information of each wheel of the unmanned vehicle obtained through wheel speed sensors. Wheel steering angle data is obtained through steering angle sensors to obtain the front wheel steering angle, used to determine the actual steering state of the vehicle and assist in path prediction.

[0029] Target environmental disturbance data is used to reflect the potential impact of the external environment on vehicle operation, including light intensity data, communication environment data, and traffic condition data. Light intensity data is acquired through light sensors, and communication environment data is obtained through onboard communication modules, such as V2X or LTE / 5G, to monitor network signal communication latency or packet loss rate to determine the reliability of data transmission. Traffic condition data is obtained through onboard cameras or roadside equipment connected to a traffic information service platform to acquire real-time road congestion information, traffic flow, or traffic accident alerts, to predict the impact of external disturbances on the driving path.

[0030] Target obstacle perception data is used to identify potential collision risks, including obstacle size, obstacle location, and passability category. The three-dimensional envelope size (length × width × height), relative position (x, y, z) of the target obstacle are detected in real time by vehicle-mounted LiDAR or millimeter-wave radar, and the passability category is marked according to the obstacle type, such as pedestrians, cones, and guardrails, such as passable, impassable, or requiring deceleration.

[0031] By collecting and integrating multi-source heterogeneous data on the target vehicle's own status, environmental disturbances, and obstacle information, the vehicle can achieve global perception of itself and the environment, thereby improving the effectiveness and reliability of dynamic obstacle avoidance decisions.

[0032] The target driving control data and target environmental disturbance data in the target multi-source heterogeneous data are jointly reconstructed to obtain the target motion state estimation result of the target unmanned vehicle.

[0033] Furthermore, the target driving control data and target environmental disturbance data in the target multi-source heterogeneous data are jointly reconstructed to obtain the target motion state estimation result of the target unmanned vehicle. This includes: processing the target driving control data and the target environmental disturbance data based on the adaptive capacitive Kalman filter method to form a measurement input; analyzing the measurement input through a preset unmanned vehicle dynamics model to obtain the target motion state estimation result; wherein, the preset unmanned vehicle dynamics model embeds a dynamic process noise covariance matrix, and the dynamic process noise covariance matrix is ​​obtained by dynamically correcting the preset process noise covariance matrix with a dynamic forgetting factor, and the dynamic forgetting factor is negatively correlated with the dynamic environmental factor.

[0034] Furthermore, the measured input is analyzed by a preset unmanned vehicle dynamics model to obtain the target motion state estimation result. This includes encoding and normalizing the light intensity data, the communication environment data, and the traffic condition data, and then using a variation weighting method to obtain the dynamic environment factor.

[0035] Specifically, the target driving control data and target environmental disturbance data collected by the target unmanned vehicle are preprocessed. The preprocessing steps include: using a time synchronization method to interpolate the data of each sensor to a unified timestamp; using a maximum-minimum normalization method to quantize each data of the target driving control data to the same numerical range; encoding and normalizing the light intensity data, communication environment data, and traffic condition data; and mapping heterogeneous data of different dimensions to the [0,1] interval. Among these, the light intensity data is normalized to the [0,1] interval by the light value measured by the light sensor according to the preset maximum light value; the communication delay in the communication environment data is mapped using an inverse S-shaped function, with a larger delay resulting in a larger normalization value; and the congestion information, traffic flow, or traffic accident warnings in the traffic condition data are encoded into numerical vectors and processed using the maximum-minimum normalization method.

[0036] After normalization, a variation-weighted method is used to weight the light intensity data, communication environment data, and traffic condition data. Variation-weighted calculation dynamically allocates weights based on the dispersion of the current data. If a certain type of environmental data fluctuates drastically within a recent window (i.e., has high variation), its weight is reduced to avoid outlier interference; conversely, its weight is increased. For example, for a normalized environmental data sequence, a sliding window is constructed using the most recent five sampling times. The variance of the sliding window is calculated. If the squared variance is greater than a preset threshold of 0.7, the data is considered to fluctuate drastically, and the corresponding weight should be reduced to minimize the impact of outliers. Conversely, if the variance is small, the data is considered stable, and its weight can be increased. For example, the weights for stable light intensity are 0.2, communication 0.4, and traffic 0.2. By multiplying the weights by the corresponding normalized data and summing them, a dynamic environmental factor DEF is formed to reflect the uncertainty of environmental disturbances on vehicle movement.

[0037] The processed target driving control data and target environmental disturbance data are used as inputs and processed by the adaptive capacitive Kalman filter method to form the measurement input for estimating the motion state of the autonomous vehicle. Adaptive capacitive Kalman filtering is a high-precision filtering algorithm for nonlinear systems. Its core principle is to construct high-dimensional integration points in the vehicle state space, such as 2n integration points (n=6), and numerically integrate the nonlinear state transition function to achieve accurate prediction and updating of the vehicle's motion state. Compared to extended Kalman filtering, it does not require the calculation of the Jacobian matrix, resulting in higher numerical stability. The adaptive nature is reflected in the fact that the process noise covariance matrix is ​​no longer a preset fixed value but dynamically changes with the environment.

[0038] To further enhance the adaptability of state estimation, a pre-defined autonomous vehicle dynamics model is introduced into the adaptive capacitive Kalman filter. This model employs a two-degree-of-freedom bicycle model, with its state vector comprising six dimensions: the vehicle's position (x, y) in the global coordinate system, heading angle, longitudinal velocity, lateral velocity, and yaw rate. This state vector simulates the autonomous vehicle's motion behavior under different input and environmental conditions. A pre-defined process noise covariance matrix (Q0) is embedded within the dynamics model to describe the inherent deviation between the model's predictions and the actual motion state, and is pre-calibrated using vehicle dynamics characteristics.

[0039] A dynamic forgetting factor is introduced to dynamically correct the preset process noise covariance matrix, resulting in a dynamic process noise covariance matrix. For example, the dynamic process noise covariance matrix Q = dynamic forgetting factor λ × preset process noise covariance matrix Q0. The dynamic forgetting factor λ is negatively correlated with the dynamic environmental factor DEF, for example, expressed as λ = e -α⋅DEFSince α>0, the larger the dynamic environmental factor, the smaller the dynamic forgetting factor λ, thus increasing the weight of the measurement values. This makes the filter more sensitive to new measurement data when external disturbances are significant, enabling it to respond quickly to environmental changes. The processed measurement input is then used to predict and update the state through a preset autonomous vehicle dynamics model, obtaining real-time motion state estimation results for the target autonomous vehicle, including key motion parameters such as global vehicle position, velocity vector, acceleration vector, heading angle, and yaw rate. This provides accurate and reliable basic data for subsequent candidate obstacle avoidance trajectory generation, collision threat assessment, and optimal obstacle avoidance trajectory planning.

[0040] In traditional Kalman filtering schemes, the process noise covariance matrix is ​​a fixed value, leading to a sharp drop in state estimation accuracy when the environment changes drastically. This, in turn, causes subsequent trajectory planning to generate dangerous trajectories based on incorrect state input. This application introduces an adaptive forgetting mechanism driven by dynamic environmental factors, enabling automatic weighted measurement of state estimation in harsh environments and maintaining the smoothness of the autonomous vehicle's dynamics model in favorable environments. This ensures that the output target motion state estimation results have high reliability across all scenarios, further improving the safety and reliability of autonomous vehicle driving control.

[0041] A set of candidate obstacle avoidance driving trajectories is generated by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and a first candidate trajectory is extracted from the set of candidate obstacle avoidance driving trajectories, wherein the first candidate trajectory corresponds to a first position uncertainty boundary.

[0042] Furthermore, a set of target candidate obstacle avoidance driving trajectories is generated by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and a first candidate trajectory is extracted from the target candidate obstacle avoidance driving trajectory set, wherein the first candidate trajectory corresponds to a first position uncertainty boundary, including: introducing a two-dimensional Gaussian distribution to determine the position uncertainty confidence interval; obtaining the historical driving logs of the first candidate trajectory and filtering out candidate historical trajectories whose trajectory similarity reaches a predetermined threshold; taking the historical trajectories whose confidence level is in the position uncertainty confidence interval from the candidate historical trajectories to form a first target historical trajectory; and using the first target historical trajectory to constitute the first position uncertainty boundary.

[0043] Specifically, based on the real-time motion state estimation results of the target unmanned vehicle and the pre-designed obstacle avoidance maneuver command set, the motion state estimation results are used as initial input conditions, and the pre-designed obstacle avoidance maneuver command set is used as constraint rules to generate a set of target candidate obstacle avoidance driving trajectories. The obstacle avoidance maneuver command set refers to multiple sets of executable trajectory actions preset for different obstacle positions and types, such as emergency stop, left or right deviation, deceleration and lane change, and emergency acceleration. Each trajectory corresponds to a specific vehicle control input sequence, such as wheel steering angle and wheel speed changes. From the set of target candidate obstacle avoidance driving trajectories, the first candidate trajectory with the minimum sum of state deviations, i.e., the highest similarity, is extracted.

[0044] To quantitatively describe the potential deviations of a vehicle during trajectory execution, a two-dimensional Gaussian distribution is introduced to represent the confidence interval for position uncertainty. This two-dimensional Gaussian distribution refers to the probability density function of a two-dimensional normal distribution describing the uncertainty of the actual position of each point on the first candidate trajectory, with the mean center as the driving plane of the autonomous vehicle (XY plane). Position uncertainty refers to the degree of deviation between the actual and estimated positions of the autonomous vehicle due to factors such as sensor noise, model errors, and environmental interference. The confidence interval is an elliptical region that covers the true position with a certain probability, such as 95%, under this two-dimensional Gaussian distribution. For example, using the estimated covariance matrix from the output target motion state estimation result (a 6×6 matrix), a 3×3 submatrix related to position and heading angle is taken, denoted as Ppos, and used as the covariance parameter Σ of the two-dimensional Gaussian distribution. At any point pi on the first candidate trajectory, the probability distribution of the actual position is N(pi,Σ), and its 95% confidence interval is an ellipse centered at pi. The direction of the major axis of the ellipse is determined by the eigenvectors of Σ, and the lengths of the major and minor axes are determined by the eigenvalues. By continuously stitching together the confidence intervals of all points on the first candidate trajectory, a position uncertainty confidence interval is formed, indicating that when the autonomous vehicle travels along the first candidate trajectory, there is a 95% probability that its true position falls within this area.

[0045] The historical driving logs of the first candidate trajectory are obtained. These logs record the trajectories executed by the autonomous vehicle under similar environmental or road conditions, including time-series data such as position, speed, and acceleration. Each historical trajectory is then extracted from the logs, and its similarity to the first candidate trajectory is calculated. Any calculated similarity is compared to a predetermined threshold, such as 0.80, and trajectories with a similarity greater than or equal to this threshold are selected as candidate historical trajectories. For each point on each candidate historical trajectory, a two-dimensional Gaussian probability density value is calculated relative to the corresponding point on the first candidate trajectory. If this Gaussian probability density value is greater than a set confidence threshold, such as the density threshold corresponding to a 95% confidence interval, the point is considered to be within the confidence interval. Trajectories where all points are within the confidence interval are then selected to form the first target historical trajectory.

[0046] All trajectories in the historical trajectory set of the first target are superimposed on a two-dimensional plane, and their outer envelope, i.e. the maximum and minimum boundaries of all trajectory points, is taken as the final first position uncertainty boundary. This indicates that in similar past scenarios, the autonomous vehicle has never exceeded this boundary in actual driving, and it is used as the current safe reference range for obstacle avoidance.

[0047] By combining real-time motion state estimation with historical trajectory data, a vehicle position uncertainty boundary can be constructed, which can effectively quantify the potential deviations and safety risks of unmanned vehicles when executing candidate trajectories, provide reliable safety constraints for dynamic obstacle avoidance decisions, and ensure the safe driving of unmanned vehicles in complex environments.

[0048] Furthermore, the historical driving logs of the first candidate trajectory are obtained, and candidate historical trajectories with trajectory similarity reaching a predetermined threshold are selected, including: extracting any historical trajectory from the historical driving logs; evaluating the trajectory similarity between the first candidate trajectory and the arbitrary historical trajectory to obtain an arbitrary similarity; if the arbitrary similarity is within the predetermined threshold, then adding the arbitrary historical trajectory to the candidate historical trajectory.

[0049] Specifically, to ensure that the autonomous vehicle can refer to past practical experience when executing candidate obstacle avoidance trajectories, an arbitrary historical trajectory is first extracted from the historical driving log, with the length of this historical trajectory being consistent with the first candidate trajectory. The mean squared error (MSE) method or dynamic time rule method is used to evaluate the trajectory similarity between the first candidate trajectory and the extracted arbitrary historical trajectory. For example, the MSE method is used, and the trajectory similarity between the two is calculated using the MSE formula; the closer the value is to 1, the more similar the trajectories are. The calculated arbitrary similarity is compared with a preset threshold, such as 0.8. If the arbitrary similarity is greater than or equal to the preset threshold, the historical trajectory is added to the candidate historical trajectory set; if it is less than the preset threshold, the historical trajectory is discarded.

[0050] By iterating through all the trajectories in the historical driving log, a complete set of candidate historical trajectories is obtained. This set not only includes historical trajectories similar to the first candidate trajectory, but also reflects the deviation range of the vehicle's actual movement under similar environments through the data characteristics of the historical trajectories.

[0051] By screening the first candidate trajectory for similarity with historical driving data, and associating the vehicle's past actual motion experience with the current candidate trajectory, the potential deviations and uncertainties in trajectory execution can be effectively quantified. This provides data support for generating safe and reliable position uncertainty boundaries, thereby improving the effectiveness and safety of autonomous vehicle dynamic obstacle avoidance.

[0052] The first candidate trajectory and the first position uncertainty boundary are used to construct a target collision avoidance envelope, and the collision threat assessment is performed by combining the target obstacle perception data to obtain the target assessment result.

[0053] Specifically, each sampling point of the first candidate trajectory is used as the center point, and a rectangular or elliptical envelope is constructed based on the lateral and longitudinal deviations defined by the position uncertainty boundary. For example, if the position of a sampling point is (xi, yi) and the position uncertainty is ±0.2m, then the collision avoidance envelope of this sampling point can be represented as a rectangular region [(xi-0.2, yi-0.2), (xi+0.2, yi+0.2)]. The envelopes of all sampling points are connected along the trajectory to form a continuous target collision avoidance envelope.

[0054] Collision threat assessment is performed by combining target obstacle perception data with the target collision avoidance envelope. If the target collision avoidance envelope and the target obstacle spatial area intersect, a collision threat is considered to exist. The threat level is further quantified, for example, by calculating the collision risk value based on the intersection area, the relative speed of the obstacle, and the distance, thus forming the target assessment result. The collision risk value = distance weight × (overlapping area of ​​the collision avoidance envelope and the obstacle / total area of ​​the collision avoidance envelope). The distance weight is inversely proportional to the distance, and the closer the distance, the greater the weight.

[0055] By constructing a target collision avoidance envelope and combining it with obstacle perception data for collision threat assessment, the risk of potential collisions between autonomous vehicles and surrounding obstacles during candidate trajectory execution can be quantified, providing precise safety constraints for dynamic obstacle avoidance and ensuring that autonomous vehicles can drive safely in complex environments.

[0056] Based on the target evaluation results, the target obstacle avoidance mode is triggered, and the optimal obstacle avoidance trajectory is determined according to the obstacle avoidance adaptation strategy under the target obstacle avoidance mode.

[0057] Furthermore, based on the target evaluation result, a target obstacle avoidance mode is triggered, and the optimal obstacle avoidance trajectory is determined according to the obstacle avoidance fitness strategy under the target obstacle avoidance mode, including: under the target obstacle avoidance mode, an obstacle avoidance fitness evaluation is performed on the first candidate trajectory to obtain a first fitness; the target candidate obstacle avoidance trajectory set is optimized with the first fitness as the target to determine the optimal obstacle avoidance trajectory; wherein, the first fitness is the normalized result of the mean trajectory similarity between each historical trajectory in the first target historical trajectory and the first candidate trajectory.

[0058] Specifically, based on the target assessment results, i.e. the collision risk assessment value of the candidate trajectory, it is determined whether to trigger the target obstacle avoidance mode. When the first candidate trajectory has a high collision threat, i.e. the collision risk value is greater than 0.3, the UAV vehicle automatically enters the target obstacle avoidance mode and activates the obstacle avoidance adaptation strategy in this target obstacle avoidance mode to comprehensively evaluate and optimize the target candidate obstacle avoidance driving trajectory set.

[0059] First, in the target obstacle avoidance mode, the obstacle avoidance fitness of the first candidate trajectory is evaluated. The average similarity between the first candidate trajectory and its historical trajectory is calculated and normalized to obtain the first fitness of the first candidate trajectory. The same obstacle avoidance fitness evaluation process is performed on each candidate trajectory in the target candidate obstacle avoidance trajectory set to obtain its own fitness. Then, with the goal of maximizing fitness, the target candidate obstacle avoidance trajectory set is optimized. The fitness values ​​of all target candidate obstacle avoidance trajectory sets are traversed, and the candidate obstacle avoidance trajectory with the highest fitness is selected as the optimal obstacle avoidance trajectory. If the trajectory corresponding to the highest fitness is the first candidate trajectory, then the optimal obstacle avoidance trajectory is the first candidate trajectory itself. If the highest fitness corresponds to another candidate trajectory ck, then the optimal obstacle avoidance trajectory is ck.

[0060] By evaluating and optimizing the obstacle avoidance adaptability of candidate trajectories, the optimal trajectory can be selected quickly and reliably from multiple potential obstacle avoidance trajectories. This enables unmanned vehicles to avoid obstacles in complex environments while maintaining stable and executable motion behavior, thereby improving the safety and reliability of unmanned vehicle driving.

[0061] The obstacle avoidance control of the target unmanned vehicle is performed based on the optimal obstacle avoidance trajectory.

[0062] Specifically, the optimal obstacle avoidance trajectory is used as the final control target for real-time obstacle avoidance control of the target unmanned vehicle. This obstacle avoidance control refers to generating control inputs for the unmanned vehicle based on the optimal trajectory, including wheel steering angle, wheel speed, and acceleration commands, enabling the unmanned vehicle to travel smoothly and safely along the optimal trajectory while avoiding collisions with obstacles in the environment. During vehicle operation, multi-source heterogeneous data from the unmanned vehicle is continuously acquired to form a closed-loop control system. If the vehicle deviates from the optimal trajectory or detects a new obstacle, the steering angle and speed can be adjusted in real time to achieve dynamic obstacle avoidance. By converting the optimal obstacle avoidance trajectory into actual control commands, the dynamic obstacle avoidance operation of the unmanned vehicle is realized, ensuring that the vehicle can avoid obstacles and drive smoothly and controllably in complex environments.

[0063] Example 2 is based on the same inventive concept as the dynamic obstacle avoidance-based unmanned vehicle driving control method in the previous examples, such as... Figure 2 As shown, this application provides an unmanned vehicle driving control system based on dynamic obstacle avoidance, wherein the unmanned vehicle driving control system based on dynamic obstacle avoidance includes:

[0064] The system comprises: a data acquisition module 11 for real-time acquisition of multi-source heterogeneous data of the target unmanned vehicle; a joint reconstruction module 12 for joint reconstruction of the target driving control data and target environmental disturbance data in the multi-source heterogeneous data to obtain the target motion state estimation result of the target unmanned vehicle; a candidate trajectory extraction module 13 for generating a target candidate obstacle avoidance driving trajectory set by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and extracting the first candidate trajectory from the target candidate obstacle avoidance driving trajectory set, wherein the first candidate trajectory corresponds to the first position uncertainty boundary; a collision assessment module 14 for constructing a target collision avoidance envelope by coordinating the first candidate trajectory and the first position uncertainty boundary, and performing a collision threat assessment by combining the target obstacle perception data to obtain the target assessment result; a trajectory determination module 15 for triggering a target obstacle avoidance mode based on the target assessment result, and determining the optimal obstacle avoidance driving trajectory according to the obstacle avoidance fitness strategy in the target obstacle avoidance mode; and an obstacle avoidance control module 16 for performing obstacle avoidance control of the target unmanned vehicle based on the optimal obstacle avoidance driving trajectory.

[0065] Furthermore, the data acquisition module 11 is also used to: include at least the target multi-source heterogeneous data, the target driving control data, the target environmental disturbance data, and the target obstacle perception data; wherein, the target driving control data includes satellite positioning data, inertial measurement data, wheel speed data, and wheel steering angle data; wherein, the target environmental disturbance data includes light intensity data, communication environment data, and traffic condition data; wherein, the target obstacle perception data includes obstacle size, obstacle location, and passability category.

[0066] Furthermore, the joint reconstruction module 12 is also used to: process the target driving control data and the target environmental disturbance data based on the adaptive capacitive Kalman filtering method to form a measurement input; analyze the measurement input through a preset unmanned vehicle dynamics model to obtain the target motion state estimation result; wherein, the preset unmanned vehicle dynamics model embeds a dynamic process noise covariance matrix, and the dynamic process noise covariance matrix is ​​obtained by dynamically correcting the preset process noise covariance matrix by a dynamic forgetting factor, and the dynamic forgetting factor is negatively correlated with the dynamic environmental factor.

[0067] Furthermore, the joint reconstruction module 12 is also used to: analyze the measurement input through a preset unmanned vehicle dynamics model to obtain the target motion state estimation result, which includes: encoding and normalizing the light intensity data, the communication environment data and the traffic condition data, and obtaining the dynamic environment factor by variation weighting.

[0068] Furthermore, the candidate trajectory extraction module 13 is also used to: introduce a two-dimensional Gaussian distribution to determine the confidence interval of the position uncertainty; obtain the historical driving log of the first candidate trajectory and filter out candidate historical trajectories whose trajectory similarity reaches a predetermined threshold; take the historical trajectories whose confidence level is in the confidence interval of the position uncertainty from the candidate historical trajectories to form the first target historical trajectory; and use the first target historical trajectory to form the first position uncertainty boundary.

[0069] Furthermore, the candidate trajectory extraction module 13 is also used to: extract any historical trajectory from the historical driving log; perform trajectory similarity evaluation on the first candidate trajectory and the arbitrary historical trajectory to obtain an arbitrary similarity; if the arbitrary similarity is within the predetermined threshold, then add the arbitrary historical trajectory to the candidate historical trajectory.

[0070] Furthermore, the trajectory determination module 15 is also used to: evaluate the obstacle avoidance fitness of the first candidate trajectory in the target obstacle avoidance mode to obtain a first fitness; optimize the target candidate obstacle avoidance driving trajectory set with the first fitness as the target to determine the optimal obstacle avoidance driving trajectory; wherein, the first fitness is the normalized result of the mean trajectory similarity between each historical trajectory in the first target historical trajectory and the first candidate trajectory.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The unmanned vehicle driving control method and specific examples based on dynamic obstacle avoidance in the aforementioned embodiment one are also applicable to the unmanned vehicle driving control system based on dynamic obstacle avoidance in this embodiment. Through the foregoing detailed description of the unmanned vehicle driving control method based on dynamic obstacle avoidance, those skilled in the art can clearly understand the unmanned vehicle driving control system based on dynamic obstacle avoidance in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0073] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for controlling the driving of an unmanned vehicle based on dynamic obstacle avoidance, characterized in that, include: Real-time acquisition of multi-source heterogeneous data of the target unmanned vehicle; The target driving control data and target environmental disturbance data in the target multi-source heterogeneous data are jointly reconstructed to obtain the target motion state estimation result of the target unmanned vehicle. A set of candidate obstacle avoidance driving trajectories for the target is generated by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and a first candidate trajectory is extracted from the set of candidate obstacle avoidance driving trajectories for the target, wherein the first candidate trajectory corresponds to a first position uncertainty boundary; The first candidate trajectory and the first position uncertainty boundary are used to construct a target collision avoidance envelope, and the collision threat assessment is performed by combining the target obstacle perception data to obtain the target assessment result; Based on the target evaluation results, the target obstacle avoidance mode is triggered, and the optimal obstacle avoidance trajectory is determined according to the obstacle avoidance adaptation strategy under the target obstacle avoidance mode. The obstacle avoidance control of the target unmanned vehicle is performed based on the optimal obstacle avoidance trajectory.

2. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 1, characterized in that, The target multi-source heterogeneous data includes at least the target driving control data, the target environmental disturbance data, and the target obstacle perception data; The target driving control data includes satellite positioning data, inertial measurement data, wheel speed data, and wheel steering angle data; The target environmental disturbance data includes light intensity data, communication environment data, and traffic condition data. The target obstacle perception data includes obstacle size, obstacle location, and passability category.

3. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 2, characterized in that, The target driving control data and target environmental disturbance data in the target multi-source heterogeneous data are jointly reconstructed to obtain the target motion state estimation result of the target unmanned vehicle, including: The target driving control data and the target environmental disturbance data are processed using an adaptive capacitive Kalman filter method to form a measurement input; The measured input is analyzed by a preset unmanned vehicle dynamics model to obtain the target motion state estimation result; The preset unmanned vehicle dynamics model embeds a dynamic process noise covariance matrix, which is obtained by dynamically correcting the preset process noise covariance matrix using a dynamic forgetting factor, and the dynamic forgetting factor is negatively correlated with dynamic environmental factors.

4. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 3, characterized in that, The measured input is analyzed by a preset unmanned vehicle dynamics model to obtain the target motion state estimation result. The preceding steps include: encoding and normalizing the light intensity data, the communication environment data, and the traffic condition data, and then obtaining the dynamic environment factor by variation weighting.

5. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 1, characterized in that, A set of candidate obstacle avoidance maneuvers is generated by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and a first candidate trajectory is extracted from the set of candidate obstacle avoidance trajectories, wherein the first candidate trajectory corresponds to a first position uncertainty boundary, including: A two-dimensional Gaussian distribution is introduced to determine the confidence interval for the location uncertainty; Obtain the historical driving logs of the first candidate trajectory, and filter out candidate historical trajectories whose trajectory similarity reaches a predetermined threshold; The first target historical trajectory is formed by selecting historical trajectories whose confidence level is within the position uncertainty confidence interval from the candidate historical trajectories. The first position uncertainty boundary is formed by the historical trajectory of the first target.

6. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 3, characterized in that, Obtain the historical driving logs of the first candidate trajectory, and filter out candidate historical trajectories whose trajectory similarity reaches a predetermined threshold, including: Extract any historical trajectory from the historical driving log; The first candidate trajectory and the arbitrary historical trajectory are evaluated for trajectory similarity to obtain an arbitrary similarity. If any similarity is within the predetermined threshold, then the arbitrary historical trajectory is added to the candidate historical trajectory.

7. The unmanned vehicle driving control method based on dynamic obstacle avoidance as described in claim 5, characterized in that, Based on the target evaluation result, a target obstacle avoidance mode is triggered, and the optimal obstacle avoidance trajectory is determined according to the obstacle avoidance adaptation strategy under the target obstacle avoidance mode, including: In the target obstacle avoidance mode, the obstacle avoidance fitness of the first candidate trajectory is evaluated to obtain the first fitness. The optimal obstacle avoidance trajectory is determined by optimizing the set of candidate obstacle avoidance trajectories with the first fitness as the objective. Wherein, the first fitness is the normalized result of the mean trajectory similarity between each historical trajectory in the first target historical trajectory and the first candidate trajectory.

8. An unmanned vehicle driving control system based on dynamic obstacle avoidance, characterized in that, The steps for implementing the unmanned vehicle driving control method based on dynamic obstacle avoidance according to any one of claims 1 to 7, wherein the unmanned vehicle driving control system based on dynamic obstacle avoidance includes: The data acquisition module is used to collect multi-source heterogeneous data of the target unmanned vehicle in real time; The joint reconstruction module is used to jointly reconstruct the target driving control data and the target environmental disturbance data in the target multi-source heterogeneous data to obtain the target motion state estimation result of the target unmanned vehicle. The candidate trajectory extraction module is used to generate a target candidate obstacle avoidance driving trajectory set by coordinating a preset obstacle avoidance maneuver command set with the target motion state estimation result, and to extract the first candidate trajectory in the target candidate obstacle avoidance driving trajectory set, wherein the first candidate trajectory corresponds to the first position uncertainty boundary; The collision assessment module is used to construct a target collision avoidance envelope by coordinating the first candidate trajectory and the first position uncertainty boundary, and to perform a collision threat assessment by combining the target obstacle perception data to obtain the target assessment result; The trajectory determination module is used to trigger the target obstacle avoidance mode based on the target evaluation result, and determine the optimal obstacle avoidance driving trajectory according to the obstacle avoidance adaptation strategy under the target obstacle avoidance mode. The obstacle avoidance control module is used to control the obstacle avoidance of the target unmanned vehicle based on the optimal obstacle avoidance driving trajectory.

Images