Two-wheeled vehicle track prediction and path deviation correction system based on reinforcement learning

Abstract

The invention discloses a two-wheeled vehicle track prediction and path deviation correction system based on reinforcement learning, and the system comprises the following modules: a data standardization module which is used for collecting and standardizing the data of a multi-source sensor; the clustering analysis module is used for performing clustering analysis on the standardized sensor data set through an improved fuzzy C-means clustering algorithm to obtain a fuzzy environment feature set; the environment perception graph generation module is used for generating an environment perception graph by calculating an Euclidean distance between clustering centers in the fuzzy environment feature set; the target trajectory prediction module is used for constructing a Laplacian matrix according to the environmental perception map and performing future trajectory prediction of the two-wheeled vehicle to obtain a target trajectory; the path deviation generation module is used for calculating path deviation; and the path correction module is used for performing path correction based on the path deviation. According to the method, the improved fuzzy C-means clustering algorithm is combined, and high-precision track prediction and path correction of the two-wheeled vehicle are achieved.

Description

Technical Field

[0001] This invention relates to the field of path planning and control technology, and in particular to a two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning. Background Technology

[0002] In the fields of intelligent transportation systems and autonomous driving, trajectory prediction and path deviation correction technologies are among the key technologies for ensuring the safe and efficient operation of autonomous driving systems. Existing trajectory prediction methods for two-wheeled vehicles are mostly based on traditional control algorithms, such as PID control and fuzzy control. These methods predict future trajectories by modeling the vehicle's motion state and utilizing its instantaneous position, velocity, and acceleration information. However, these methods have poor adaptability in complex environments, especially when facing dynamic obstacles and complex traffic scenarios. Traditional algorithms often struggle to guarantee the accuracy and stability of path prediction. Therefore, how to perform high-precision trajectory prediction in complex environments and correct deviations in a timely manner has become a major challenge in autonomous driving technology.

[0003] Most existing path deviation correction methods rely on local path-based control strategies or fixed models for correction. These methods often assume minimal environmental changes or relatively regular vehicle paths, neglecting dynamic changes in complex traffic environments. For example, traditional path correction methods typically adjust based on the difference between the vehicle's current trajectory and the target trajectory, lacking comprehensive perception and adaptation to the vehicle's surrounding environment. In situations with obstacles, sudden traffic conditions, or environmental changes, traditional methods may lead to delayed or excessively large path corrections, impacting vehicle safety and efficiency. Many existing methods rely heavily on environmental perception during path planning and correction. For instance, traditional path prediction often calculates based solely on vehicle dynamics (such as speed, acceleration, and steering angle), ignoring the influence of external environmental factors like road conditions, traffic signals, and obstacles. Even advanced algorithms that incorporate sensor data often only perform simple fusion processing, failing to fully utilize the correlation and complexity between multi-source data and the potential for complex patterns in the environment. This results in existing methods having weak responsiveness to unexpected situations and lacking dynamic adjustment capabilities in practical applications.

[0004] Therefore, how to provide a two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning. This invention achieves high-precision two-wheeled vehicle trajectory prediction and path deviation correction by fusing multi-source sensor data, an improved fuzzy C-means clustering algorithm, an environmental perception map, and a Laplace matrix. This system can perceive complex environments in real time and dynamically correct deviations, improving the driving safety, stability, and navigation capabilities of autonomous two-wheeled vehicles in dynamic traffic environments, overcoming the shortcomings of existing technologies such as lagging path correction, insufficient accuracy, and poor adaptability.

[0006] The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to an embodiment of the present invention includes the following modules:

[0007] The data standardization module is used to collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure.

[0008] The clustering analysis module is used to perform clustering analysis on the standardized sensor dataset using an improved fuzzy C-means clustering algorithm, determine the initial cluster centers, calculate the membership degree of each initial cluster center, and obtain the fuzzy environment feature set.

[0009] The environment perception map generation module is used to generate an environment perception map by calculating the Euclidean distance between the cluster centers of fuzzy environment feature sets.

[0010] The target trajectory prediction module is used to calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory.

[0011] The path deviation generation module is used to compare the vehicle's current position with the target trajectory and calculate the path deviation.

[0012] The path correction module is used to correct paths based on path deviations.

[0013] The two-wheeled vehicle trajectory prediction and path deviation correction method based on reinforcement learning according to embodiments of the present invention includes the following steps:

[0014] Step 1: Collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure;

[0015] Step 2: Perform cluster analysis on the standardized sensor dataset using the improved fuzzy C-means clustering algorithm to determine the initial cluster centers and calculate the membership degree of each initial cluster center to obtain the fuzzy environment feature set;

[0016] Step 3: Generate an environment perception map by calculating the Euclidean distance between cluster centers based on the fuzzy environment feature set;

[0017] Step 4: Calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory;

[0018] Step 5: Compare the vehicle's current position with the target trajectory and calculate the path deviation;

[0019] Step 6: Correct the path based on the path deviation.

[0020] Optionally, the multi-source sensor data includes GPS data, inertial measurement unit data, lidar point cloud data, and road image data; the inertial measurement unit data includes acceleration data, angular velocity data, and steering angle data; the road image feature data includes road conditions, obstacle distribution, and traffic sign information of the surrounding environment; the standardization step includes outlier removal, missing value imputation, timestamp alignment, and normalization processing for different types of data to obtain a standardized sensor dataset with a unified structure.

[0021] Optionally, the improved fuzzy C-means clustering algorithm is specifically as follows:

[0022] Based on standardized sensor datasets, vehicle status information and road image feature data are organized into a multi-dimensional data matrix;

[0023] The vehicle's status information includes GPS data, inertial measurement unit data, and lidar point cloud data; each column of the multidimensional data matrix represents a sample point; and each sample point consists of data from different sensors at the same time.

[0024] C sample points are randomly selected from the standardized sensor dataset as the initial cluster centers, where C is the preset number of clusters;

[0025] Calculate the Euclidean distance between each sample point and each initial cluster center;

[0026] Based on the calculated Euclidean distance from the sample points to the initial cluster centers, the membership degree of each sample point to each initial cluster center is calculated, and a membership degree matrix is ​​generated.

[0027] The membership degree is the ratio of the Euclidean distance of each sample point to an initial cluster center to the sum of the Euclidean distances to all cluster centers, representing the degree of membership of the sample point to the initial cluster center.

[0028] Based on the calculated membership matrix, the position of each cluster center is updated. The position of the cluster center is the weighted average position of the sample points in the same column of the membership matrix, and the weight is the corresponding membership degree.

[0029] Calculate the change in the membership matrix based on the updated cluster centers:

[0030] If the difference between the membership matrix before the update and the value before the update is less than the set threshold, it indicates that the clustering result has converged and the iteration stops.

[0031] If the difference between the membership matrix before the update and the original value is greater than or equal to the set threshold, then update the membership matrix and cluster centers until the clustering results converge.

[0032] Based on the final clustering results, each sample point is classified into different clusters according to the degree of membership, and the average feature of each cluster is extracted to obtain the fuzzy environment feature set.

[0033] Optionally, step three specifically includes:

[0034] Each cluster in the fuzzy environmental feature set is used as a node in the environmental perception map, and each cluster represents a specific environmental feature.

[0035] For each cluster, calculate the Euclidean distance between the cluster centers, where the Euclidean distance represents the similarity between clusters;

[0036] Based on the calculated Euclidean distance between cluster centers;

[0037] If the Euclidean distance between two clusters is less than a preset distance threshold, then a connection edge is established between the corresponding two nodes.

[0038] The weight of the connecting edge is the Euclidean distance between the corresponding cluster centers.

[0039] Optionally, step four specifically involves:

[0040] An adjacency matrix is ​​constructed based on the nodes and edges in the environment perception graph. The adjacency matrix is ​​a symmetric matrix. If there is no edge connecting two nodes, the value of the corresponding element in the adjacency matrix is ​​0.

[0041] If there is an edge connecting two nodes, then the element value in the corresponding adjacency matrix is ​​the reciprocal of the weight of the connecting edge;

[0042] Calculate the degree matrix of the environment perception map, where each element in the degree matrix is ​​the degree of the corresponding node, and the degree is the number of connected edges of the node;

[0043] The Laplace matrix is ​​obtained by subtracting the degree matrix from the adjacency matrix.

[0044] Choose the top left corner node of the Laplacian matrix as the starting node, and initialize the shortest path value of all nodes to infinity, with the shortest path value of the starting node being 0.

[0045] Create a collection of visited nodes to store the nodes for which the shortest path has been found, and add the starting node to the collection of visited nodes;

[0046] From all unvisited nodes, select the node with the smallest shortest path value and denote it as the current node;

[0047] For each neighboring node of the current node, calculate the path length through the current node and compare it with the known shortest path value;

[0048] If the path value from the current node to the adjacent node plus the shortest path value of the current node is less than the shortest path value of the adjacent node, then update the shortest path value of the adjacent node, where the path value is the element value of the Laplace matrix.

[0049] Mark the current node as visited and add it to the set of visited nodes. Continue iterating until all nodes have been visited, and obtain the future target trajectory of the two-wheeled vehicle starting from the starting node.

[0050] Optionally, step five specifically includes:

[0051] The path deviation is calculated by comparing the vehicle's current position with the target trajectory. The path deviation includes distance deviation and angle deviation.

[0052] The distance deviation is the spatial distance between the vehicle's current position and the nearest point on the target trajectory, and the angle deviation is the angle between the vehicle's steering angle and the direction of the target trajectory.

[0053] Optionally, step six specifically includes:

[0054] If the sum of the distance deviation and the angle deviation is less than the preset deviation threshold, the target trajectory does not need to be corrected.

[0055] If the sum of the distance deviation and the angle deviation is greater than or equal to a preset deviation threshold, the target trajectory is corrected, and the correction includes:

[0056] Adjust the vehicle's steering angle according to the angle deviation until the vehicle's direction of travel is consistent with the direction of the target trajectory;

[0057] The speed is adjusted according to the spatial distance between the vehicle and the target trajectory until the spatial distance between the vehicle and the nearest point on the target trajectory reaches the minimum value.

[0058] The beneficial effects of this invention are:

[0059] This invention overcomes the problems of insufficient accuracy, poor adaptability, and lagging path correction in existing technologies by combining an improved fuzzy C-means clustering algorithm, environmental perception map generation, and Laplace matrix calculation. Traditional trajectory prediction and path correction methods often lack in-depth perception of complex environments and dynamic changes, leading to untimely path correction and potentially significant vehicle deviations or departures from the target trajectory. This invention, however, integrates multi-source sensor data, including GPS, IMU, LiDAR, and road image data, performs standardized processing, and uses an improved fuzzy C-means clustering algorithm to perform cluster analysis, extracting more accurate and comprehensive environmental features. These features, through the construction of an environmental perception map, provide a solid data foundation for subsequent trajectory prediction and path correction, enabling the system to perceive environmental changes in real time and make dynamic adjustments based on the current situation.

[0060] By calculating the Laplace matrix and applying the Dijkstra algorithm for target trajectory prediction, this invention can accurately predict the vehicle's expected driving path in complex environments, rather than relying solely on traditional control strategies. This method effectively integrates environmental perception information, ensuring that trajectory prediction considers not only the vehicle's current dynamic state but also external environmental factors such as road conditions and obstacle distribution, thus improving the accuracy and robustness of trajectory prediction. Regarding path deviation calculation and correction, this invention uses a comparison between the vehicle's current position and the target trajectory to calculate the path deviation in real time and determines whether correction is needed based on the deviation result. When the path deviation exceeds a set threshold, the system adjusts the vehicle's steering angle and speed to ensure the vehicle quickly returns to the predetermined trajectory. Compared to traditional methods, this invention can correct the path more accurately and promptly, avoiding the lag in path correction and greatly improving the driving safety and stability of two-wheeled vehicles in dynamic and complex environments. This invention achieves more efficient and accurate two-wheeled vehicle trajectory prediction and path correction, solving the problems of untimely, insufficient accuracy, and poor adaptability in existing technologies for path correction in complex environments, and significantly improving the autonomous navigation capability and driving efficiency of autonomous two-wheeled vehicles in complex traffic environments. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is a schematic diagram of the structure of the two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning proposed in this invention;

[0063] Figure 2 This is an overall flowchart of the reinforcement learning-based two-wheeled vehicle trajectory prediction and path deviation correction method proposed in this invention. Detailed Implementation

[0064] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0065] refer to Figure 1 A two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning includes the following modules:

[0066] The data standardization module is used to collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure.

[0067] The clustering analysis module is used to perform clustering analysis on the standardized sensor dataset using an improved fuzzy C-means clustering algorithm, determine the initial cluster centers, calculate the membership degree of each initial cluster center, and obtain the fuzzy environment feature set.

[0068] The environment perception map generation module is used to generate an environment perception map by calculating the Euclidean distance between the cluster centers of fuzzy environment feature sets.

[0069] The target trajectory prediction module is used to calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory.

[0070] The path deviation generation module is used to compare the vehicle's current position with the target trajectory and calculate the path deviation.

[0071] The path correction module is used to correct paths based on path deviations.

[0072] This step, by combining multi-source sensor data and reinforcement learning, realizes a highly efficient two-wheeled vehicle trajectory prediction and path deviation correction system. Through a data standardization module, the system can uniformly process multi-source sensor data from GPS, IMU, LiDAR, and road images, ensuring data quality. An improved fuzzy C-means clustering algorithm accurately extracts environmental features and generates an environmental perception map for comprehensive road condition analysis. Laplace matrix and Dijkstra's algorithm are used for trajectory prediction, ensuring the accuracy of the predicted path. The path deviation calculation module can compare the current position with the target trajectory in real time, quickly generating a path correction scheme, thereby significantly improving the driving accuracy and safety of two-wheeled vehicles in complex environments, solving the problems of lagging and insufficient accuracy in path correction in existing technologies.

[0073] refer to Figure 2 A reinforcement learning-based method for two-wheeled vehicle trajectory prediction and path deviation correction includes the following steps:

[0074] Step 1: Collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure;

[0075] Step 2: Perform cluster analysis on the standardized sensor dataset using the improved fuzzy C-means clustering algorithm to determine the initial cluster centers and calculate the membership degree of each initial cluster center to obtain the fuzzy environment feature set;

[0076] Step 3: Generate an environment perception map by calculating the Euclidean distance between cluster centers based on the fuzzy environment feature set;

[0077] Step 4: Calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory;

[0078] Step 5: Compare the vehicle's current position with the target trajectory and calculate the path deviation;

[0079] Step 6: Correct the path based on the path deviation.

[0080] This process achieves efficient prediction and path correction of two-wheeled vehicle trajectories through a series of steps. First, the system standardizes data from different sensors using a multi-source sensor data acquisition module, ensuring data uniformity and accuracy. Next, an improved fuzzy C-means clustering algorithm performs cluster analysis on the standardized data, accurately extracting environmental features and generating a fuzzy environmental feature set. Based on these features, the system calculates the Euclidean distance between cluster centers, generates an environmental perception map, and uses the Laplace matrix to predict future trajectories. By calculating the deviation between the vehicle's current position and the target trajectory, the system can assess and correct path deviations in real time, ensuring stable vehicle operation in dynamic and complex environments. This method overcomes the limitations of traditional path correction techniques in complex scenarios, improving the safety and reliability of autonomous driving.

[0081] In this embodiment, the multi-source sensor data includes GPS data, inertial measurement unit data, lidar point cloud data, and road image data; the inertial measurement unit data (IMU data) includes data from accelerometer and gyroscope sensors; the accelerometer measures the vehicle's acceleration in three directions (X, Y, and Z axes); the gyroscope measures the vehicle's rotational angular velocity in three axes (pitch, roll, and yaw). IMU data is used to calculate vehicle acceleration, angular velocity, and steering angle, providing dynamic information for trajectory prediction. The lidar point cloud data is used to measure distances around the vehicle. LiDAR emits laser beams and receives reflected laser signals to construct 3D point cloud data, helping to build a 3D model of the environment and providing obstacle detection, road boundary, and surrounding object location information. The road image data is road image information acquired by cameras, including lane lines, traffic signs, obstacles, road surface conditions, and other visual information. Camera data helps with visual recognition, providing detection of static and dynamic objects in the environment. The standardization steps include outlier removal, missing value imputation, timestamp alignment, and normalization of different data types to obtain a sensor dataset with a unified structure.

[0082] This step improves the accuracy and reliability of two-wheeled vehicle trajectory prediction and path correction by comprehensively utilizing multi-source sensor data. The system integrates GPS data, inertial measurement unit (IMU) data, LiDAR point cloud data, and road image data to comprehensively capture the vehicle's motion state and surrounding environment. IMU data provides information on vehicle acceleration, angular velocity, and steering angle, helping to more accurately predict the vehicle's dynamic behavior. Road image feature data includes road conditions, obstacle distribution, and traffic sign information, ensuring comprehensive environmental perception. Through data standardization processing, including outlier removal, missing value imputation, timestamp alignment, and normalization, the system ensures data consistency and accuracy, providing a reliable foundation for subsequent trajectory prediction and path correction. This multi-source data fusion and standardization technology significantly enhances the system's adaptability and robustness in complex environments.

[0083] In this embodiment, the improved fuzzy C-means clustering algorithm is specifically as follows:

[0084] Based on standardized sensor datasets, vehicle status information and road image feature data are organized into a multi-dimensional data matrix;

[0085] The vehicle's status information includes GPS data, inertial measurement unit data, and lidar point cloud data; each column of the multidimensional data matrix represents a sample point; and each sample point consists of data from different sensors at the same time.

[0086] C sample points are randomly selected from the standardized sensor dataset as the initial cluster centers, where C is the preset number of clusters;

[0087] Calculate the Euclidean distance between each sample point and each initial cluster center;

[0088] Based on the calculated Euclidean distance from the sample points to the initial cluster centers, the membership degree of each sample point to each initial cluster center is calculated, and a membership degree matrix is ​​generated.

[0089] The membership degree is the ratio of the Euclidean distance of each sample point to an initial cluster center to the sum of the Euclidean distances to all cluster centers, representing the degree of membership of the sample point to the initial cluster center.

[0090] Based on the calculated membership matrix, the position of each cluster center is updated. The position of the cluster center is the weighted average position of the sample points in the same column of the membership matrix, and the weight is the corresponding membership degree.

[0091] Calculate the change in the membership matrix based on the updated cluster centers:

[0092] If the difference between the membership matrix before the update and the value before the update is less than the set threshold, it indicates that the clustering result has converged and the iteration stops.

[0093] If the difference between the membership matrix before the update and the original value is greater than or equal to the set threshold, then update the membership matrix and cluster centers until the clustering results converge.

[0094] Based on the final clustering results, each sample point is classified into different clusters according to the degree of membership, and the average feature of each cluster is extracted to obtain the fuzzy environment feature set.

[0095] This step significantly improves the accuracy and adaptability of two-wheeled vehicle trajectory prediction and path correction through an improved fuzzy C-means clustering algorithm. Based on a standardized multi-source sensor dataset, the algorithm comprehensively considers vehicle state information and road image feature data, organizing them into a multi-dimensional data matrix for effective cluster analysis. By randomly selecting cluster centers and calculating the Euclidean distance between sample points and initial cluster centers, the algorithm accurately determines the membership degree of each sample point to different cluster centers. Iterative optimization of cluster center positions until the membership matrix converges ensures the stability and reliability of the clustering results. Ultimately, the fuzzy environmental feature set generated by the algorithm provides high-precision environmental perception information, supporting subsequent trajectory prediction and path correction. This method not only improves the accuracy of trajectory prediction but also effectively enhances the system's adaptability to complex environments, overcoming the insufficient accuracy and instability problems of traditional clustering algorithms.

[0096] In this embodiment, step three specifically includes:

[0097] Each cluster in the fuzzy environmental feature set is used as a node in the environmental perception map, and each cluster represents a specific environmental feature.

[0098] For each cluster, calculate the Euclidean distance between the cluster centers, where the Euclidean distance represents the similarity between clusters;

[0099] Based on the calculated Euclidean distance between cluster centers;

[0100] If the Euclidean distance between two clusters is less than a preset distance threshold, then a connection edge is established between the corresponding two nodes.

[0101] The weight of the connecting edge is the Euclidean distance between the corresponding cluster centers.

[0102] This step innovatively improves the accuracy and robustness of two-wheeled vehicle trajectory prediction and path correction by using each cluster in the fuzzy environmental feature set as a node in the environmental perception graph. Each cluster represents a specific environmental feature, such as road type, obstacle distribution, or road surface condition, ensuring comprehensive environmental perception. By calculating the Euclidean distance between cluster centers, the system can accurately assess the similarity between clusters, thus providing more refined environmental feature information for path correction. If the Euclidean distance between two clusters is less than a preset distance threshold, the system expresses their similarity relationship by establishing a connecting edge, with the weight of the edge proportional to the Euclidean distance, further enhancing the accuracy and detail of the graph structure. This method provides strong environmental support for path correction, making path correction for two-wheeled vehicles more accurate and real-time in complex environments, effectively solving the problems of insufficient environmental perception and untimely path correction in traditional methods.

[0103] In this embodiment, step four specifically includes:

[0104] An adjacency matrix is ​​constructed based on the nodes and edges in the environment perception graph. The adjacency matrix is ​​a symmetric matrix. If there is no edge connecting two nodes, the value of the corresponding element in the adjacency matrix is ​​0.

[0105] If there is an edge connecting two nodes, then the element value in the corresponding adjacency matrix is ​​the reciprocal of the weight of the connecting edge;

[0106] Calculate the degree matrix of the environment perception map, where each element in the degree matrix is ​​the degree of the corresponding node, and the degree is the number of connected edges of the node;

[0107] The Laplace matrix is ​​obtained by subtracting the degree matrix from the adjacency matrix.

[0108] Choose the top left corner node of the Laplacian matrix as the starting node, and initialize the shortest path value of all nodes to infinity, with the shortest path value of the starting node being 0.

[0109] Create a collection of visited nodes to store the nodes for which the shortest path has been found, and add the starting node to the collection of visited nodes;

[0110] From all unvisited nodes, select the node with the smallest shortest path value and denote it as the current node;

[0111] For each neighboring node of the current node, calculate the path length through the current node and compare it with the known shortest path value;

[0112] If the path value from the current node to the adjacent node plus the shortest path value of the current node is less than the shortest path value of the adjacent node, then update the shortest path value of the adjacent node, where the path value is the element value of the Laplace matrix.

[0113] Mark the current node as visited and add it to the set of visited nodes. Continue iterating until all nodes have been visited, and obtain the future target trajectory of the two-wheeled vehicle starting from the starting node.

[0114] This step significantly improves the accuracy and flexibility of two-wheeled vehicle trajectory prediction and path correction by constructing a Laplace matrix based on the environmental perception map. By calculating the adjacency matrix, the system can clearly express the similarity and relationships between nodes, ensuring the accurate transmission of environmental information. The construction of the Laplace matrix further strengthens the structural features of the graph, making trajectory prediction more accurate and reliable. The system can efficiently calculate the shortest path, i.e., the future target trajectory. This method can not only accurately predict the vehicle's driving route but also adjust the path in real time according to environmental changes, ensuring that the vehicle always travels along the predetermined trajectory in dynamic environments. By optimizing the path selection and update mechanism, this invention effectively solves the problems of inaccurate path prediction and lagging path correction in traditional methods, improving the safety and stability of autonomous two-wheeled vehicles in complex environments.

[0115] In this embodiment, step five specifically includes:

[0116] The path deviation is calculated by comparing the vehicle's current position with the target trajectory. The path deviation includes distance deviation and angle deviation.

[0117] The distance deviation is the spatial distance between the vehicle's current position and the nearest point on the target trajectory, and the angle deviation is the angle between the vehicle's steering angle and the direction of the target trajectory.

[0118] This step, by accurately calculating path deviation, effectively improves the real-time performance and accuracy of two-wheeled vehicle trajectory prediction and path correction. By comparing the vehicle's current position with the target trajectory, the system can accurately assess the deviation between the current driving path and the predetermined trajectory. Distance deviation is obtained by calculating the spatial distance between the vehicle's current position and the nearest point on the target trajectory, reflecting the actual degree of deviation between the vehicle and the target trajectory in real time. Angle deviation is calculated by comparing the angle between the vehicle's current steering angle and the direction of the target trajectory, ensuring that the vehicle's driving direction always remains consistent with the target trajectory. This method provides a precise basis for subsequent path correction, effectively solving the problems of lag or inaccuracy in path correction in traditional methods, and improving the adaptability and driving safety of autonomous driving systems in complex environments.

[0119] In this embodiment, step six specifically includes:

[0120] If the sum of the distance deviation and the angle deviation is less than the preset deviation threshold, the target trajectory does not need to be corrected.

[0121] If the sum of the distance deviation and the angle deviation is greater than or equal to a preset deviation threshold, the target trajectory is corrected, and the correction includes:

[0122] Adjust the vehicle's steering angle according to the angle deviation until the vehicle's direction of travel is consistent with the direction of the target trajectory;

[0123] The speed is adjusted according to the spatial distance between the vehicle and the target trajectory until the spatial distance between the vehicle and the nearest point on the target trajectory reaches the minimum value.

[0124] This step significantly improves the trajectory tracking accuracy of the two-wheeled vehicle by dynamically evaluating path deviation and making real-time corrections based on the deviation. When the sum of distance and angle deviations is less than a preset deviation threshold, the system considers the vehicle to be on the target trajectory and requires no correction, thus avoiding unnecessary adjustments and improving system efficiency. If the deviation exceeds the set threshold, the system activates the path correction mechanism, adjusting the vehicle's steering angle according to the angle deviation to align its direction of travel with the target trajectory. Simultaneously, the system adjusts the vehicle's speed to minimize the distance between the vehicle and the target trajectory, achieving high-precision path correction. This mechanism effectively responds to changes in the dynamic environment, ensuring that the autonomous two-wheeled vehicle maintains accurate path tracking in complex road conditions, enhancing the safety and stability of vehicle operation.

[0125] Example 1:

[0126] To verify the feasibility of this invention in practice, it was applied to an urban road environment scenario. Under complex and dynamically changing road conditions, the system was used to achieve high-precision and high-efficiency trajectory correction. Two-wheeled vehicles, as a lightweight and flexible mode of transportation, are widely used for short-distance urban travel. With the continuous development of autonomous driving technology, improving the path planning and correction capabilities of autonomous two-wheeled vehicles in complex environments has become a key issue in enhancing their intelligence and safety. By employing the technology of this invention, this embodiment can effectively improve the trajectory prediction accuracy and path correction capability of two-wheeled vehicles in actual driving, solving the problems of lagging path correction, insufficient accuracy, and poor dynamic adaptability in traditional methods.

[0127] In this application scenario, the autonomous two-wheeled vehicle integrates multiple sensors, including GPS, inertial measurement unit (IMU), LiDAR, and cameras, to comprehensively perceive the vehicle's status and surrounding environment. GPS data provides the vehicle's location information, IMU data provides the vehicle's acceleration, angular velocity, and steering angle, while LiDAR and cameras are used to monitor surrounding obstacles, road conditions, and traffic signs in real time. All the data from these sensors is processed through a data standardization module to ensure the uniformity and accuracy of the sensor data. The standardization process includes outlier removal, missing value imputation, timestamp alignment, and normalization to eliminate differences between data from different sensors and improve data consistency.

[0128] The processed data is then transmitted to a clustering analysis module, which performs clustering analysis on the standardized sensor dataset based on an improved fuzzy C-means clustering algorithm. Each cluster represents a specific environmental feature, such as different types of roads, obstacle distribution, and road surface conditions. Through this clustering analysis, the system can extract fuzzy environmental features related to the target trajectory, providing rich environmental perception data for subsequent trajectory prediction.

[0129] After generating the fuzzy environment feature set, the system generates an environment perception map by calculating the Euclidean distance between cluster centers. This environment perception map constructs a graphical model reflecting road surface conditions, obstacle distribution, and road topology by treating each cluster as a node and establishing connections based on the similarity between clusters. The environment perception map provides a comprehensive environment model, providing an accurate foundation for subsequent trajectory prediction and path correction. Using the environment perception map, the system further calculates the Laplace matrix and predicts the target trajectory. The Laplace matrix combines the relationship between the adjacency matrix and the degree matrix, accurately expressing the relationships between nodes in the environment. This matrix can be used to calculate the shortest path from the starting node to the target node. This shortest path is the expected travel trajectory of the two-wheeled vehicle over a period of time. Based on the structural information of the graph, the optimal path from the current vehicle position to the target trajectory can be calculated efficiently, ensuring that the vehicle can travel along the predetermined path in complex environments.

[0130] After obtaining the target trajectory, the system compares the vehicle's current position with the target trajectory using a path deviation generation module to calculate the path deviation. Path deviation includes distance deviation and angle deviation. Distance deviation represents the spatial distance between the vehicle's current position and the nearest point on the target trajectory, while angle deviation is the angle between the vehicle's steering angle and the direction of the target trajectory. By calculating these deviations, the system can monitor in real time whether the vehicle deviates from the target trajectory and decide whether to perform path correction based on the deviation. When the system detects that the path deviation exceeds a preset threshold, it triggers the path correction module to perform real-time path adjustment. Path correction includes both steering angle adjustment and speed adjustment. Based on the calculated angle deviation, the system adjusts the vehicle's steering angle to align the vehicle's direction with the target trajectory. Simultaneously, the system adjusts the vehicle's speed based on the distance deviation to reduce the spatial distance to the nearest point on the target trajectory, thereby ensuring the vehicle returns to the target trajectory as quickly as possible. This path correction mechanism can respond to dynamically changing environments in real time, ensuring that the two-wheeled vehicle always travels along the target trajectory.

[0131] To verify the effectiveness of this invention, the system was tested in an urban road environment. The test section included various complex road conditions, such as curves, intersections, and obstacles. Real-time acquisition and processing of sensor data ensured the accuracy of trajectory prediction and correction. During the test, the system was able to accurately predict the vehicle's target trajectory and correct the path in real time based on the deviation between the vehicle's current position and the target trajectory. Compared with traditional path correction methods, the system of this invention not only improves path correction accuracy but also significantly shortens the correction response time.

[0132] To further verify the effectiveness of this invention, we compared it with three traditional path correction algorithms: a path correction algorithm based on PID control, a path correction algorithm based on model predictive control (MPC), and a path correction method based on the shortest path algorithm. Experiments were conducted in the same urban road environment, and the experimental data are as follows:

[0133] Table 1 Comparison of Experimental Data

[0134] Data item PID control algorithm MPC algorithm Shortest path algorithm Method of the invention Path deviation (m) 3.8 3.2 3.0 1.2 Angular deviation (°) 18 12 14 5 Correction response time (s) 12 8 10 3 Path correction accuracy 80% 85% 82% 98% Average speed (km / h) 16 15 17 18

[0135] As shown in Table 1, the PID-based path correction algorithm performs poorly in terms of path deviation, angle deviation, and correction response time, with a correction accuracy of only 80%. While path correction methods based on MPC and the shortest path algorithm show some improvement, they still cannot provide sufficiently high correction accuracy in complex environments, and their correction response time is relatively long. In contrast, the method of this invention achieves a path correction accuracy of 98% and a correction response time of only 3 seconds, significantly outperforming traditional methods. These experimental data demonstrate that the method of this invention not only has a significant advantage in path correction accuracy but also excels in correction response time and dynamic adaptability. This system can perceive changes in the environment in real time and quickly correct deviations, ensuring that the two-wheeled vehicle can drive accurately and safely in complex urban road environments, greatly improving the performance and safety of autonomous driving. This fully demonstrates the application effect of the reinforcement learning-based two-wheeled vehicle trajectory prediction and path deviation correction system in complex dynamic environments, overcoming the limitations of traditional path correction methods in real-world environments and providing a more efficient and accurate path prediction and correction solution. Through the application of this system, autonomous two-wheeled vehicles can drive stably in complex urban environments, accurately predict trajectories, and correct deviations in a timely manner, thereby greatly improving driving safety and stability. The system's real-time performance, high precision, and adaptability make it a promising candidate for applications in intelligent transportation and autonomous driving.

[0136] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning, characterized in that, Includes the following modules: The data standardization module is used to collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure. The clustering analysis module is used to perform clustering analysis on the standardized sensor dataset using an improved fuzzy C-means clustering algorithm, determine the initial cluster centers, calculate the membership degree of each initial cluster center, and obtain the fuzzy environment feature set. The environment perception map generation module is used to generate an environment perception map by calculating the Euclidean distance between the cluster centers of fuzzy environment feature sets. The target trajectory prediction module is used to calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory. The path deviation generation module is used to compare the vehicle's current position with the target trajectory and calculate the path deviation. The path correction module is used to correct paths based on path deviations.

2. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 1, characterized in that, The modules are connected in the following way: Step 1: Collect multi-source sensor data and standardize the multi-source sensor data to obtain a standardized sensor dataset with a unified structure; Step 2: Perform cluster analysis on the standardized sensor dataset using the improved fuzzy C-means clustering algorithm to determine the initial cluster centers and calculate the membership degree of each initial cluster center to obtain the fuzzy environment feature set; Step 3: Generate an environment perception map by calculating the Euclidean distance between cluster centers based on the fuzzy environment feature set; Step 4: Calculate the adjacency matrix and degree matrix of the environmental perception map, construct the Laplacian matrix, and predict the future trajectory of the two-wheeled vehicle to obtain the target trajectory; Step 5: Compare the vehicle's current position with the target trajectory and calculate the path deviation; Step 6: Correct the path based on the path deviation.

3. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, The multi-source sensor data includes GPS data, inertial measurement unit data, lidar point cloud data, and road image data; the inertial measurement unit data includes acceleration data, angular velocity data, and steering angle data; the road image feature data includes the road conditions of the surrounding environment, obstacle distribution, and traffic sign information; the standardization steps include outlier removal, missing value imputation, timestamp alignment, and normalization processing for different types of data to obtain a standardized sensor dataset with a unified structure.

4. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, The improved fuzzy C-means clustering algorithm is specifically as follows: Based on standardized sensor datasets, vehicle status information and road image feature data are organized into a multi-dimensional data matrix; The vehicle's status information includes GPS data, inertial measurement unit data, and lidar point cloud data; each column of the multidimensional data matrix represents a sample point; and each sample point consists of data from different sensors at the same time. C sample points are randomly selected from the standardized sensor dataset as the initial cluster centers, where C is the preset number of clusters; Calculate the Euclidean distance between each sample point and each initial cluster center; Based on the calculated Euclidean distance from the sample points to the initial cluster centers, the membership degree of each sample point to each initial cluster center is calculated, and a membership degree matrix is ​​generated. The membership degree is the ratio of the Euclidean distance of each sample point to an initial cluster center to the sum of the Euclidean distances to all cluster centers, representing the degree of membership of the sample point to the initial cluster center. Based on the calculated membership matrix, the position of each cluster center is updated. The position of the cluster center is the weighted average position of the sample points in the same column of the membership matrix, and the weight is the corresponding membership degree. Calculate the change in the membership matrix based on the updated cluster centers: If the difference between the membership matrix before the update and the value before the update is less than the set threshold, it indicates that the clustering result has converged and the iteration stops. If the difference between the membership matrix and the original membership matrix is ​​greater than or equal to the set threshold, then update the membership matrix and cluster centers until the clustering results converge. Based on the final clustering results, each sample point is classified into different clusters according to the degree of membership, and the average feature of each cluster is extracted to obtain the fuzzy environment feature set.

5. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, Step three specifically involves: Each cluster in the fuzzy environmental feature set is used as a node in the environmental perception map, and each cluster represents a specific environmental feature. For each cluster, calculate the Euclidean distance between the cluster centers, where the Euclidean distance represents the similarity between clusters; Based on the calculated Euclidean distance between cluster centers; If the Euclidean distance between two clusters is less than a preset distance threshold, then a connection edge is established between the corresponding two nodes. The weight of the connecting edge is the Euclidean distance between the corresponding cluster centers.

6. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, Step four specifically involves: An adjacency matrix is ​​constructed based on the nodes and edges in the environment perception graph. The adjacency matrix is ​​a symmetric matrix. If there is no edge connecting two nodes, the value of the corresponding element in the adjacency matrix is ​​0. If there is an edge connecting two nodes, then the element value in the corresponding adjacency matrix is ​​the reciprocal of the weight of the connecting edge; Calculate the degree matrix of the environment perception map, where each element in the degree matrix is ​​the degree of the corresponding node, and the degree is the number of connected edges of the node; The Laplace matrix is ​​obtained by subtracting the degree matrix from the adjacency matrix. Choose the top left corner node of the Laplacian matrix as the starting node, and initialize the shortest path value of all nodes to infinity, with the shortest path value of the starting node being 0. Create a collection of visited nodes to store the nodes for which the shortest path has been found, and add the starting node to the collection of visited nodes; From all unvisited nodes, select the node with the smallest shortest path value and denote it as the current node; For each neighboring node of the current node, calculate the path length through the current node and compare it with the known shortest path value; If the path value from the current node to the adjacent node plus the shortest path value of the current node is less than the shortest path value of the adjacent node, then update the shortest path value of the adjacent node, where the path value is the element value of the Laplace matrix. Mark the current node as visited and add it to the set of visited nodes. Continue iterating until all nodes have been visited, and obtain the future target trajectory of the two-wheeled vehicle starting from the starting node.

7. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, Step five specifically involves: The path deviation is calculated by comparing the vehicle's current position with the target trajectory. The path deviation includes distance deviation and angle deviation. The distance deviation is the spatial distance between the vehicle's current position and the nearest point on the target trajectory, and the angle deviation is the angle between the vehicle's steering angle and the direction of the target trajectory.

8. The two-wheeled vehicle trajectory prediction and path deviation correction system based on reinforcement learning according to claim 2, characterized in that, Step six specifically involves: If the sum of the distance deviation and the angle deviation is less than the preset deviation threshold, the target trajectory does not need to be corrected. If the sum of the distance deviation and the angle deviation is greater than or equal to a preset deviation threshold, the target trajectory is corrected, and the correction includes: Adjust the vehicle's steering angle according to the angle deviation until the vehicle's direction of travel is consistent with the direction of the target trajectory; The speed is adjusted according to the spatial distance between the vehicle and the target trajectory until the spatial distance between the vehicle and the nearest point on the target trajectory reaches the minimum value.

Images