A Local Trajectory Planning Method and System Based on a Stochastic Model

Abstract

This invention belongs to the field of autonomous driving technology, specifically relating to a local trajectory planning method and system based on a stochastic model. The method includes: acquiring trajectory data of traffic participants as model training data based on an existing traffic database; training and evaluating a long short-term memory (LSTM) network model based on the training data to obtain the optimal LSM network model; acquiring trajectory data of dynamic obstacles within a fixed time interval, and predicting the motion trajectory of dynamic obstacles within future time intervals based on the optimal LSM network model; determining potential collision points based on the motion trajectory of dynamic obstacles and the vehicle's own trajectory, and constructing a set of regional sampling probabilities by combining the sampling probabilities of different regions within the sampling area; performing local path search using a fast expanding random tree algorithm based on a stochastic model to generate local paths; and using dynamic programming to perform velocity planning on the generated local paths to complete the vehicle's local trajectory planning task.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a local trajectory planning method and system based on a stochastic model. Background Technology

[0002] Path planning methods are rapidly developing as a core technology in the field of autonomous driving. They aim to calculate and optimize the safest and most efficient driving path in real time in complex and ever-changing road environments by fusing advanced algorithms with sensors.

[0003] Existing path planning methods struggle to quickly and accurately generate safe local paths in complex road environments, especially those with multiple dynamic obstacles. Most existing methods rely on offline computation or simple heuristics, making them difficult to apply to highly dynamic environments. This is particularly true when the environment has numerous obstacles and interwoven trajectories, further complicating path solving and hindering the rapid generation of local paths. Existing methods widely utilize A... * Algorithms such as A, Dijkstra's algorithm, and Rapid Expanding Random Tree (RRT) are suitable for static environments, but adaptable to complex planning tasks in dynamic environments. Among them, graph search-based algorithms (such as A) are particularly suitable for complex planning tasks. * While algorithms like the one used by Dijkstra and others can find the optimal path, their high computational complexity makes them unsuitable for real-time requirements. Furthermore, while Long Short-Term Memory (LSTM) networks, a special type of recurrent neural network, excel in processing time-series data and are widely applicable to trajectory prediction and motion planning, relying solely on LSTM networks for path planning is insufficient for fast and efficient path finding in complex dynamic environments. Summary of the Invention

[0004] This invention provides a local trajectory planning method and system based on a stochastic model. By combining a long short-term memory network model and a fast extended random tree algorithm improved based on a stochastic model, and by incorporating a planning strategy that separates path and velocity in time and space, the complexity of path planning problems in complex dynamic scenarios is reduced, the solution speed of path planning is improved, and the real-time performance of the algorithm is enhanced. At the same time, it can achieve accurate prediction of dynamic obstacle trajectories and fast local path planning in complex road environments.

[0005] A local trajectory planning method based on a stochastic model includes the following steps:

[0006] Based on the established traffic database, the trajectory data of traffic participants are obtained as model training data;

[0007] Based on the model training data, using kThe cross-validation mechanism is used to train and evaluate the long short-term memory network model to obtain the optimal long short-term memory network model.

[0008] Acquire trajectory data of dynamic obstacles within a fixed time interval, and predict the motion trajectory of dynamic obstacles within future time intervals based on the best long short-term memory network model;

[0009] Potential collision points are determined based on the motion trajectories of dynamic obstacles and the vehicle's own trajectory, and a set of regional sampling probabilities is constructed by combining the sampling probabilities of different areas within the sampling area.

[0010] Based on the region sampling probability set, a fast extended random tree algorithm based on a randomness model is used to perform local path search and generate local paths.

[0011] Dynamic programming is used to perform velocity planning on the generated local path, thus completing the task of local trajectory planning for the vehicle.

[0012] By combining a long short-term memory network model and a fast extended random tree algorithm improved based on a stochastic model, and by incorporating a planning strategy that separates path and velocity in time and space, the complexity of path planning problems in complex dynamic scenarios is reduced, the solution speed of path planning is improved, and the real-time performance of the algorithm is enhanced. At the same time, it can achieve accurate prediction of dynamic obstacle trajectories and fast local path planning in complex road environments.

[0013] Furthermore, based on the established traffic database, the process of obtaining the trajectory data of traffic participants as model training data specifically includes the following steps:

[0014] Based on the established traffic database, extract the trajectory data of traffic participants;

[0015] Preprocess the trajectory data of traffic participants;

[0016] Feature selection is performed on the preprocessed trajectory data of traffic participants, and this data is used as training data for the model.

[0017] Furthermore, based on the model training data, using k The process of using cross-validation to train and evaluate a long short-term memory (LSTM) network model to obtain the optimal LSM network model includes the following steps:

[0018] Based on the model training data, it is divided into multiple matching training sets and validation sets;

[0019] Multiple training sets are sequentially input into a long short-term memory network model to train and obtain the root mean square error of the corresponding validation set.

[0020] Construct by combining all the obtained root mean square errors kCross-validation mechanism is used to evaluate the performance of long short-term memory network models. k Cross-validation values;

[0021] Repeat the training multiple times and select the smallest result. k The long short-term memory network model corresponding to the cross-validation value is taken as the optimal long short-term memory network model.

[0022] Furthermore, the process of acquiring trajectory data of dynamic obstacles within a fixed time interval and predicting the motion trajectory of dynamic obstacles within future time intervals based on the optimal long short-term memory network model specifically includes the following steps:

[0023] Acquire trajectory data of multiple dynamic obstacles on the road within a fixed time interval;

[0024] Feature selection is performed on the trajectory data of multiple pre-processed dynamic obstacles within a fixed time interval, and this data is used as the trajectory data.

[0025] The trajectory data of multiple dynamic obstacles selected by features within a fixed time interval are input into the optimal long short-term memory network model, and the model is trained to output the motion trajectory of the dynamic obstacles within a future time interval.

[0026] Furthermore, the process of determining potential collision points based on the motion trajectories of dynamic obstacles and the vehicle's own trajectory, and constructing a set of regional sampling probabilities by combining the sampling probabilities of different areas within the sampling area, specifically includes the following steps:

[0027] Based on the motion trajectories of multiple dynamic obstacles and the vehicle's own trajectory, the distance between the vehicle and each dynamic obstacle is predicted and calculated, and the minimum collision time point is obtained by solving the problem.

[0028] Based on the minimum collision time point, determine the specific location where the collision occurred and mark it as the potential collision point of the vehicle;

[0029] Based on the potential collision points of the vehicle, the sampling area is determined, and the initial sampling probability within the sampling area is obtained;

[0030] The sampling probability within a single sampling region is updated based on the trajectory of a single dynamic obstacle.

[0031] The sampling probability within the intersection region formed by the trajectories of multiple dynamic obstacles is updated.

[0032] Based on the unoccupied areas of the trajectory of the dynamic obstacle, update the sampling probability within the occupied areas;

[0033] By combining the sampling probabilities of single sampling regions, intersection regions, and unoccupied regions, a set of region sampling probabilities is constructed.

[0034] Furthermore, based on the region sampling probability set, an improved fast expanding random tree algorithm is used for local path search. The process of generating local paths specifically includes the following steps:

[0035] Based on the sampling region, a starting point and a target point are defined, and the starting point is used as the root node of the initial random tree;

[0036] A random point is generated in the configuration space based on the set of regional sampling probabilities.

[0037] Based on a random tree, find the node closest to the random point as the neighboring node;

[0038] Calculate the distance from a neighboring point to a random point, and expand along this direction by a step size to generate a new node such that the new node satisfies the maximum step size limit;

[0039] Collision detection is performed on the path from the neighboring point to the new node to ensure that the path meets the collision-free requirement;

[0040] Perform distance detection on the distance from the target point to the new node to ensure that the distance meets the preset distance threshold;

[0041] Repeat the above steps until the maximum number of iterations is reached, backtracking from the target point to the starting point to generate a complete local path.

[0042] Furthermore, dynamic programming is used to perform velocity planning on the generated local path. The process of solving the vehicle's local trajectory planning task specifically includes the following steps:

[0043] Smooth the generated local paths;

[0044] Construct the ST graph and mark the areas occupied by the trajectories of dynamic obstacles;

[0045] Based on the area occupied by the trajectory of dynamic obstacles, a dynamic programming method is used to plan the speed of the local path, solve for the ST path curve with the minimum cost, and complete the local trajectory planning task of the vehicle.

[0046] A system for a local trajectory planning method based on a stochastic model includes:

[0047] The data acquisition module is used to acquire trajectory data of traffic participants as model training data based on the established traffic database.

[0048] The model building module is used to build upon the model training data. k The cross-validation mechanism is used to train and evaluate the long short-term memory network model to obtain the optimal long short-term memory network model.

[0049] The trajectory prediction module is used to acquire trajectory data of dynamic obstacles within a fixed time interval and predict the motion trajectory of dynamic obstacles within a future time interval based on the best long short-term memory network model.

[0050] The probability calculation module is used to determine potential collision points based on the motion trajectory of dynamic obstacles and the trajectory of the vehicle itself, and to construct a set of regional sampling probabilities by combining the sampling probabilities of different areas within the sampling area.

[0051] The local path generation module is used to perform local path search based on the region sampling probability set and an improved fast expanding random tree algorithm to generate local paths;

[0052] The local trajectory planning module is used to perform velocity planning on the generated local path using dynamic programming methods, thereby completing the local trajectory planning task for the vehicle.

[0053] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described above.

[0054] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0055] The beneficial effects of this invention are as follows:

[0056] This invention combines a long short-term memory network model and a fast extended random tree algorithm improved based on a stochastic model, along with a planning strategy that separates path and velocity in time and space. This reduces the complexity of path planning problems in complex dynamic scenarios, improves the solution speed of path planning, and enhances the real-time performance of the algorithm. At the same time, it can achieve accurate prediction of dynamic obstacle trajectories and fast local path planning in complex road environments. Attached Figure Description

[0057] Figure 1 This is a flowchart of the present invention;

[0058] Figure 2 This is a flowchart of the method of the present invention;

[0059] Figure 3 This is a schematic diagram of the system structure of the present invention;

[0060] Figure 4 This is a schematic diagram of the computer device in this invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0063] In addition, specific details are provided in the following description to facilitate a thorough understanding of the examples, and those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0064] Example 1

[0065] Figure 1 This paper presents a local trajectory planning method based on a stochastic model. By combining a long short-term memory network model and a fast extended random tree algorithm improved based on a stochastic model, along with a planning strategy that separates path and velocity in time and space, the complexity of path planning problems in complex dynamic scenarios is reduced, the solution speed of path planning is improved, and the real-time performance of the algorithm is enhanced. Simultaneously, it can achieve both accurate prediction of dynamic obstacle trajectories and fast local path planning in complex road environments. Specifically, it includes the following steps:

[0066] S1: Based on the established traffic database, obtain the trajectory data of traffic participants as model training data;

[0067] S11: Extract trajectory data of traffic participants based on the established traffic database;

[0068] S12: Preprocess the trajectory data of traffic participants;

[0069] In this embodiment, the preprocessing of the trajectory data of traffic participants includes missing value processing, outlier processing, and standardization processing to ensure data quality.

[0070] S13: Select features from the preprocessed trajectory data of traffic participants and use it as training data for the model.

[0071] The expression for feature selection is as follows:

[0072] ;

[0073] In the formula, Indicates the first Data characteristics of a dynamic obstacle at a certain moment; Indicates the first Individual road information features; Indicates the first Category feature values ​​of dynamic obstacles; Indicates the first The coordinates of a dynamic obstacle along the X-axis; Indicates the first The coordinates of a dynamic obstacle along the Y-axis; Indicates time; Indicates the first The speed of a dynamic obstacle; Indicates the first The acceleration of a dynamic obstacle; Indicates the first The acceleration of a dynamic obstacle.

[0074] S2: Based on model training data, using k The cross-validation mechanism is used to train and evaluate the long short-term memory network model to obtain the optimal long short-term memory network model.

[0075] S21: Based on the model training data, divide it into multiple matching training sets and validation sets;

[0076] In this embodiment, the model training data is divided into k A subset is selected, and one subset is verified in turn, followed by the remaining subsets. k -1 subsets are used as the corresponding training sets to form multiple sets of matching training and validation sets.

[0077] S22: Input multiple training sets sequentially into the Long Short-Term Memory network model to train and obtain the root mean square error of the corresponding validation set. ;

[0078] S23: Construct by combining all the obtained root mean square errors k Cross-validation mechanism is used to evaluate the performance of long short-term memory network models. k Cross-validation value ;

[0079] in, kThe expression for the folded cross-validation mechanism is:

[0080] ;

[0081] In the formula, This represents a performance evaluation metric for long short-term memory network models. Indicates the first The root mean square error corresponding to each data point; , This indicates the total number of data sets.

[0082] S24: Repeat the training multiple times and select the smallest result. k The long short-term memory network model corresponding to the cross-validation value is taken as the optimal long short-term memory network model.

[0083] S3: Acquire trajectory data of dynamic obstacles within a fixed time interval, and predict the motion trajectory of dynamic obstacles within future time intervals based on the best long short-term memory network model;

[0084] S31: Acquire trajectory data of multiple dynamic obstacles on the road within a fixed time interval;

[0085] S32: Select features from the trajectory data of multiple pre-processed dynamic obstacles within a fixed time interval and use them as trajectory data;

[0086] It should be noted that the preprocessing and feature selection methods are the same as those used for processing the model training data.

[0087] S33: Input the trajectory data of multiple dynamic obstacles selected by features within a fixed time interval into the optimal long short-term memory network model, and train to output the motion trajectory of the dynamic obstacles within the future time interval.

[0088] Within the future time interval, the trajectory of any set of dynamic obstacles is represented as follows:

[0089] ;

[0090] ;

[0091] In the formula, Indicates the first A dynamic obstacle regarding time The coordinate vector; Indicates the first A dynamic obstacle regarding time The trajectory fitting function; Indicates the first The area occupied by the trajectory of a dynamic obstacle;

[0092] Wherein, assuming no environmental factors, the trajectory of a dynamic obstacle within a future time interval is represented as:

[0093] ;

[0094] In the formula, This indicates the time-dependent dynamic obstacle, unaffected by environmental factors. The coordinate vector; This indicates the time-dependent dynamic obstacle, unaffected by environmental factors. The trajectory fitting function.

[0095] S4: Determine potential collision points based on the motion trajectory of dynamic obstacles and the trajectory of the vehicle itself, and construct a set of regional sampling probabilities by combining the sampling probabilities of different regions within the sampling area;

[0096] S41: Based on the motion trajectories of multiple dynamic obstacles and the vehicle's own trajectory, predict and calculate the distance between the vehicle and each dynamic obstacle, and solve for the minimum collision time point.

[0097] The expression for calculating the distance between the vehicle and the dynamic obstacle is as follows:

[0098] ;

[0099] In the formula, Indicates the relationship between the vehicle and dynamic obstacles with respect to time. The distance; Indicates the collision threshold;

[0100] S42: Determine the specific location where a dynamic obstacle collision occurs based on the minimum collision time point. And mark them as potential collision points for the vehicle;

[0101] S43: Based on the potential collision points of the vehicle, determine the sampling area and obtain the initial sampling probability within the sampling area;

[0102] The initial sampling probability is calculated using the following expression:

[0103] ;

[0104] In the formula, Indicates the sampling area Initial sampling probability within; Indicates the area of ​​the sampling region;

[0105] S44: Based on the trajectory of a single dynamic obstacle, update the sampling probability within the single sampling region;

[0106] In this embodiment, each dynamic obstacle In the sampling area Each has a corresponding trajectory area. and the maximum potential collision point In the trajectory region In this model, the sampling probability model exhibits a characteristic of radiating outwards from the maximum potential collision point. That is, the closer a location is to the maximum probability collision point, the lower its sampling probability. Therefore, an inverse Gaussian distribution function is used to update the sampling probability within a single sampling region, and its update expression is:

[0107] ;

[0108] In the formula, Indicates the first digit within the updated single-sample region. The sampling probability of a dynamic obstacle about the collision point; Indicates the first Normalization coefficients corresponding to each dynamic obstacle; Indicates standard deviation; This represents the coordinates of the potential collision point along the X-axis. Indicates the first The coordinates of a dynamic obstacle along the Y-axis;

[0109] S45: Update the sampling probability within the intersection region formed by the trajectories of multiple dynamic obstacles;

[0110] In this embodiment, each dynamic obstacle moves independently and does not interfere with each other in the spatiotemporal dimension. Their trajectory regions intersect in the spatial dimension, forming an intersection region. When the trajectories of multiple dynamic obstacles intersect to form an intersection region, the sampling probability within the intersection region is updated using a probability-weighted average of the trajectory regions of each dynamic obstacle. The update expression is as follows:

[0111] + + ;

[0112] In the formula, Indicates the first The trajectory area of ​​a dynamic obstacle , No. The trajectory area of ​​a dynamic obstacle , No. The trajectory area of ​​a dynamic obstacle The sampling probability within the intersection region; , , These respectively represent the trajectory region Trajectory Area Trajectory Area The corresponding weight coefficients that were sampled; , , They represent the first The first dynamic obstacle, the first The first dynamic obstacle, the first The sampling probability of each dynamic obstacle;

[0113] S46: Based on the unoccupied areas of the trajectory of the dynamic obstacle, update the sampling probability within the occupied area;

[0114] The expression for updating the sampling probability within the unoccupied area is as follows:

[0115] ;

[0116] In the formula, This represents the sampling probability within the unoccupied area of ​​a dynamically moving obstacle; This represents the area of ​​the region not occupied by a dynamic obstacle. This represents the sum of sampling probabilities for the area occupied by the trajectory of a dynamic obstacle.

[0117] S47: Construct a set of region sampling probabilities by combining the sampling probabilities of single sampling regions, intersection regions, and unoccupied regions. .

[0118] S5: Based on the regional sampling probability set, a fast expanding random tree algorithm based on a randomness model is used to perform local path search and generate local paths;

[0119] S51: Define the starting point based on the sampling area. and target point and the starting point As the root node of the initial random tree;

[0120] S52: Generate a random point in the configuration space based on the region sampling probability set. ;

[0121] S53: Finding points at random distances based on a random tree. The nearest node is considered a neighboring node. ;

[0122] S54: Calculate from neighboring points To random point The distance is calculated, and a new node is generated by expanding along this direction by one step. This ensures that the new node meets the maximum step size limit;

[0123] In this embodiment, when the maximum step size limit is met, the random point is... As a new node When the maximum step size limit is not met, then the neighboring points are... Along towards random point Move one step in the direction and generate a new node. ;

[0124] S55: For neighboring points To the new node Collision detection is performed on the path to ensure that the path meets the collision-free requirement;

[0125] In this embodiment, when the collision-free requirement is not met, random points are regenerated. .

[0126] S56: From the target point To the new node Distance detection is performed to ensure that the distance meets the preset distance threshold;

[0127] In this embodiment, if the distance does not meet the preset distance threshold, a new random point is generated. .

[0128] S57: Repeat the above steps until the maximum number of iterations is reached, starting from the target point. Backtracking to the starting point This generates a complete local path.

[0129] S6: Use dynamic programming to perform velocity planning on the generated local path and solve the local trajectory planning task for the vehicle.

[0130] S61: Smooth the generated local paths;

[0131] In this embodiment, a Bézier curve is used to smooth the generated local path, and its expression is:

[0132] ;

[0133] In the formula, This represents the local path in time that has been smoothed using a Bézier curve. Indicates time, ; The path point number representing the generated local path. , This represents the total number of path points in the generated local path; This represents the path points of the generated local path.

[0134] S62: Construct the ST graph and mark the areas occupied by the dynamic obstacle trajectories;

[0135] S63: Based on the area occupied by the trajectory of dynamic obstacles, dynamic programming is used to perform velocity planning on the local path, and the ST path curve with the minimum cost is obtained to complete the local trajectory planning task of the vehicle.

[0136] The cost function expression for dynamic programming is:

[0137] ;

[0138] In the formula, This represents the cost function value of dynamic programming; Indicates the travel time between adjacent waypoints; Indicates path error; Indicates the first The speed of each path point; Indicates the first The speed of each path point; Indicates the first The acceleration of each path point; Indicates the desired speed; , , , , This represents the weighting coefficients of each term in the cost function.

[0139] In this embodiment, if the solution fails, it means that the generated local trajectory path of the vehicle is not feasible, and it is necessary to regenerate the local path and perform speed planning.

[0140] Example 2

[0141] like Figure 2 As shown, this embodiment provides a local trajectory planning method based on a stochastic model, which specifically includes the following steps:

[0142] T1: Based on the established traffic database, obtain the trajectory data of traffic participants as model training data;

[0143] T11: Extract trajectory data of traffic participants based on the established traffic database;

[0144] T12: Preprocessing trajectory data of traffic participants;

[0145] T13: Feature selection is performed on the preprocessed trajectory data of traffic participants, and this data is used as training data for the model.

[0146] T2: Based on model training data, using k The cross-validation mechanism is used to train and evaluate the long short-term memory network model to obtain the optimal long short-term memory network model.

[0147] T21: Based on the model training data, it is divided into multiple matching training sets and validation sets;

[0148] T22: Input multiple training sets sequentially into the Long Short-Term Memory (LSTM) network model to train and obtain the root mean square error of the corresponding validation set;

[0149] T23: Construct by combining all the obtained root mean square errors k Cross-validation mechanism is used to evaluate the performance of the Long Short-Term Memory (LSTM) network model. k Cross-validation values;

[0150] T24: Repeat the training multiple times and select the smallest result. k The Long Short-Term Memory (LSTM) network model corresponding to the cross-validation value is taken as the optimal LSTM network model.

[0151] T3: Acquire trajectory data of dynamic obstacles within a fixed time interval, and predict the motion trajectory of dynamic obstacles within future time intervals based on the best long short-term memory network model;

[0152] T31: Real-time acquisition of trajectory data of multiple dynamic obstacles on the road within a fixed time interval;

[0153] T32: Feature selection is performed on the trajectory data of multiple pre-processed dynamic obstacles within a fixed time interval, and the results are used as trajectory data.

[0154] T33: Input the trajectory data of multiple dynamic obstacles selected by features within a fixed time interval into the optimal long short-term memory network model, and train it to output the motion trajectory of the dynamic obstacles within the future time interval.

[0155] T4: Determine potential collision points based on the motion trajectory of dynamic obstacles and the vehicle's own trajectory, and construct a set of regional sampling probabilities by combining the sampling probabilities of different areas within the sampling area;

[0156] T41: Based on the motion trajectories of multiple dynamic obstacles and the vehicle's own trajectory, predict and calculate the distance between the vehicle and each dynamic obstacle, and solve for the minimum collision time point.

[0157] T42: Based on the minimum collision time point, determine the specific location where a dynamic obstacle collision occurs and mark it as a potential collision point for the vehicle;

[0158] T43: Based on the potential collision points of the vehicle, determine the sampling area and obtain the initial sampling probability within the sampling area;

[0159] T44: Based on the trajectory of a single dynamic obstacle, update the sampling probability within the single sampling region;

[0160] T45: Based on the intersection region formed by the trajectories of multiple dynamic obstacles, update the sampling probability within the intersection region;

[0161] T46: Based on the unoccupied areas of the trajectory of the dynamic obstacle, update the sampling probability within the occupied area;

[0162] T47: Construct a set of region sampling probabilities by combining the sampling probabilities of single sampling regions, intersection regions, and unoccupied regions. .

[0163] T5: Based on the regional sampling probability set, the Fast Extended Random Tree (RRT) algorithm, which is improved based on the stochastic model, is used to perform local path search and generate local paths;

[0164] T51: Define the starting point based on the sampling area. and target point and the starting point As the root node of the initial random tree;

[0165] T52: Generates a random point in the configuration space based on the region sampling probability set. ;

[0166] T53: Finding points at random distances based on a random tree. The nearest node is considered a neighboring node. ;

[0167] T54: Calculate from neighboring points To random point The distance is calculated, and a new node is generated by expanding along this direction by one step. Determine whether the new node meets the maximum step size limit;

[0168] T541: When the new node satisfies the maximum step size limit, make the random point As a new node Execute T55;

[0169] T542: When the maximum step size limit is not met, then make the neighboring points... Along towards random point Move one step in the direction and generate a new node. Execute T55;

[0170] T55: For points from neighboring points To the new node Collision detection is performed on the path to determine whether the path meets the collision-free requirement;

[0171] T551: When the path meets the collision-free requirement, that is, the path does not have any collisions, execute T56;

[0172] T552: When the path does not meet the collision-free requirement, i.e., the path has collisions, return to T52, i.e., regenerate random points. .

[0173] T56: From the target point To the new node The distance is detected to determine whether the distance meets the preset distance threshold.

[0174] T561: When the distance meets the preset distance threshold, complete the path planning for that point and return the path;

[0175] T562: If the distance does not meet the preset distance threshold, return to T52, that is, regenerate random points. .

[0176] T57: Repeat the above steps until the maximum number of iterations is reached or the target point is successfully found. From the target point Backtracking to the starting point This generates a complete local path.

[0177] T6: Use dynamic programming to perform velocity planning on the generated local path and solve the local trajectory planning task for the vehicle.

[0178] T61: Construct an ST graph based on the generated local paths and mark the areas occupied by dynamic obstacle trajectories;

[0179] T62: Determine whether the generated local path occupies the trajectory of dynamic obstacles;

[0180] T621: When occupying a dynamic obstacle trajectory, use dynamic programming to perform speed planning on the local path, solve the problem, and determine whether the speed solution is successful.

[0181] T6211: When the velocity is successfully solved, the vehicle local trajectory planning task is completed;

[0182] T6212: Returns to T5 when the speed solution fails;

[0183] T622: When not occupying the dynamic obstacle trajectory, smooth the generated local path and assign speed characteristics to complete the vehicle local trajectory planning task.

[0184] Example 3

[0185] Based on the same design concept, such as Figure 3 As shown, this embodiment provides a local trajectory planning system based on a stochastic model, including a data acquisition module, a model building module, a trajectory prediction module, a probability calculation module, a local path generation module, and a local trajectory planning module.

[0186] Specifically, the data acquisition module is used to acquire the trajectory data of traffic participants as model training data based on the established traffic database.

[0187] Specifically, the model building module is used to build upon the model training data. k The cross-validation mechanism is used to train and evaluate the long short-term memory network model to obtain the optimal long short-term memory network model.

[0188] Specifically, the trajectory prediction module is used to acquire trajectory data of dynamic obstacles within a fixed time interval and predict the movement trajectory of dynamic obstacles within a future time interval based on the best long short-term memory network model.

[0189] Specifically, the probability calculation module is used to determine potential collision points based on the movement trajectory of dynamic obstacles and the trajectory of the vehicle itself, and to construct a set of regional sampling probabilities by combining the sampling probabilities of different areas within the sampling area.

[0190] Specifically, the local path generation module is used to perform local path search based on the set of regional sampling probabilities and an improved fast expanding random tree algorithm to generate local paths;

[0191] Specifically, the local trajectory planning module is used to perform velocity planning on the generated local path using dynamic programming methods, thereby solving the local trajectory planning task for the vehicle.

[0192] Example 4

[0193] Based on the same technical concept, embodiments of this application also provide a computer device, including a memory 1 and a processor 2, such as... Figure 4 As shown, the memory 1 stores a computer program, and the processor 2 executes the computer program to implement any of the methods described above.

[0194] The memory 1 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 1 can be an internal storage unit of a local trajectory planning system based on a stochastic model, such as a hard disk. In other embodiments, the memory 1 can also be an external storage device of the local trajectory planning system based on a stochastic model, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Furthermore, the memory 1 can include both internal storage units and external storage devices of the local trajectory planning system based on a stochastic model. The memory 1 can be used not only to store application software and various types of data installed in the local trajectory planning system based on a stochastic model, such as the code of the local trajectory planning system program, but also to temporarily store data that has been output or will be output.

[0195] In some embodiments, processor 2 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 1 or process data, such as executing a local trajectory planning system program based on a stochastic model.

[0196] The present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method described in the above-described method embodiments. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0197] The computer program product of the application page content refresh method provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the steps of the method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0198] The present invention also discloses a computer program that, when executed by a processor, implements any of the methods described in the foregoing embodiments. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0199] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0200] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0201] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0202] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0203] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0204] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0205] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0206] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0207] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A local trajectory planning method based on a stochastic model, characterized in that, The method comprises the following steps: Based on the constructed traffic database, the trajectory data of the traffic participants is obtained as the model training data; Based on the model training data, adopt k The long short-term memory network model is trained by using a fold cross-validation mechanism, and an optimal long short-term memory network model is obtained. Obtain the trajectory data of the dynamic obstacles in a fixed time interval, and predict the motion trajectory of the dynamic obstacles in a future time interval based on the best long short-term memory network model; Determine the potential collision points according to the motion trajectory of the dynamic obstacles and the vehicle's own trajectory, and combine the sampling probability of different regions in the sampling region to construct a region sampling probability set, including: Based on the motion trajectory of the multiple dynamic obstacles and the vehicle's own trajectory, the distance between the vehicle and each dynamic obstacle is predicted and calculated, and the minimum collision time point is solved and obtained; Based on the minimum collision time point, the specific position of the collision occurrence is determined, and is marked as the potential collision point of the vehicle; Based on the potential collision point of the vehicle, the sampling region is determined, and the initial sampling probability in the sampling region is obtained; Based on the single sampling region formed by the trajectory of a single dynamic obstacle, the sampling probability in the single sampling region is updated using the inverse Gaussian distribution function; Based on the intersection region formed by the trajectories of multiple dynamic obstacles, the sampling probability in the intersection region is updated using the probability weighting of the trajectory region of each dynamic obstacle; Based on the region occupied by the trajectory of the dynamic obstacle, the sampling probability in the unoccupied region is updated; the sampling probability update expression in the unoccupied region is: ； wherein, represents the sampling probability within the unoccupied region of the dynamic obstacle; represents the area of the unoccupied region of the dynamic obstacle; represents the sum of the sampling probabilities of the occupied region of the dynamic obstacle trajectory; Combine the sampling probabilities of the single sampling region, the intersection region and the unoccupied region to construct a region sampling probability set; Based on the region sampling probability set, a local path is generated by using a fast expanding random tree algorithm improved based on a randomness model for local path search; The local path generated is subjected to speed planning by using a dynamic programming method, and the vehicle local trajectory planning task is solved.

2. The local trajectory planning method based on a random model according to claim 1, wherein, Based on the constructed traffic database, the trajectory data of the traffic participants is obtained as the model training data, which comprises the following steps: Based on the constructed traffic database, the trajectory data of the traffic participants is extracted; The trajectory data of the traffic participants is preprocessed; The preprocessed trajectory data of the traffic participants is subjected to feature selection, and is used as the model training data.

3. The local trajectory planning method based on a random model according to claim 1, wherein, Based on the model training data, adopt k The process of training and evaluating the long short-term memory network model based on the fold cross-validation mechanism to obtain the best long short-term memory network model includes the following steps: Based on the model training data, a plurality of matched training sets and verification sets are divided; A plurality of training sets are sequentially input into a long short-term memory network model, and the root mean square error of the corresponding verification set is obtained by training; The root mean square error is obtained k The performance of the long short-term memory network model is evaluated by using the fold cross-validation mechanism, and the k Fold cross-validation value Repeat the training several times and select the smallest k The long short-term memory network model corresponding to the minimum cross-validation value is selected as the optimal long short-term memory network model.

4. The local trajectory planning method based on a random model according to claim 1, wherein, The trajectory data of the dynamic obstacles in a fixed time interval is obtained, and the motion trajectory of the dynamic obstacles in a future time interval is predicted based on the best long short-term memory network model, which comprises the following steps: Obtain the trajectory data of a plurality of dynamic obstacles on the road in a fixed time interval; The preprocessed trajectory data of the plurality of dynamic obstacles in a fixed time interval is subjected to feature selection, and is used as the trajectory data; The trajectory data of the plurality of dynamic obstacles in a fixed time interval is input into the best long short-term memory network model, and the motion trajectory of the dynamic obstacles in a future time interval is output by training.

5. The local trajectory planning method based on a random model according to claim 1, wherein, Based on the region sampling probability set, a local path is generated by using an improved fast expanding random tree algorithm for local path search, which comprises the following steps: Based on the sampling region, define the start point and the target point, and take the start point as the root node of the initialized random tree; Based on the set of region sampling probabilities, generate a random point in the configuration space; Based on the random tree, find the nearest node to the random point as the neighboring point; Calculate the distance from the neighboring point to the random point, and expand a step in this direction to generate a new node, so that the new node meets the maximum step length limit; Perform collision detection on the path from the neighboring point to the new node, so that the path meets the collision-free requirement; Perform distance detection on the distance from the target point to the new node, so that the distance meets the preset distance threshold; Repeat the above steps until the maximum number of iterations is reached, backtrack from the target point to the start point, and generate a complete local path.

6. The local trajectory planning method based on a random model according to claim 1, wherein, The process of solving the vehicle local trajectory planning task by adopting the dynamic programming method for speed planning of the generated local path includes the following steps: Smooth the generated local path; Construct an ST graph and mark the region occupied by the dynamic obstacle trajectory; Based on the region occupied by the dynamic obstacle trajectory, adopt the dynamic programming method for speed planning of the local path to obtain the ST path curve with the minimum cost, and complete the vehicle local trajectory planning task.

7. A system using the local trajectory planning method based on a randomness model according to claim 1, characterized in that, It includes: A data acquisition module for acquiring trajectory data of traffic participants as model training data based on a constructed traffic database; a model construction module configured to employ, based on model training data, k a fold cross-validation mechanism to train and evaluate the long short-term memory network model, and obtain an optimal long short-term memory network model; A trajectory prediction module for acquiring trajectory data of dynamic obstacles within a fixed time interval and predicting the motion trajectory of dynamic obstacles within a future time interval based on a best long short-term memory network model; A probability calculation module for determining potential collision points according to the motion trajectory of the dynamic obstacle and the vehicle's own trajectory, and constructing a set of region sampling probabilities in combination with the sampling probabilities of different regions in the sampling region; A local path generation module for generating a local path by searching for a local path based on the set of region sampling probabilities using an improved fast expanding random tree algorithm; A local trajectory planning module for solving the vehicle local trajectory planning task by adopting the dynamic programming method for speed planning of the generated local path.

8. An electronic device, comprising: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method of any one of claims 1-6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images